Matcha-TTS-PL v2: fav2 voice distilled from VoxCPM2, English words, fav2 vocoder

.gitattributes

@@ -33,3 +33,19 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+samples/e01_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e05_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e05_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e06_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e06_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e07_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e07_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e09_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e09_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e10_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e10_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e11_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e11_v2.wav filter=lfs diff=lfs merge=lfs -text
+samples/e12_teacher.wav filter=lfs diff=lfs merge=lfs -text
+samples/e12_v2.wav filter=lfs diff=lfs merge=lfs -text
+vocoder/generator_fav2 filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,428 @@
+Attribution-ShareAlike 4.0 International
+
+=======================================================================
+
+Creative Commons Corporation ("Creative Commons") is not a law firm and
+does not provide legal services or legal advice. Distribution of
+Creative Commons public licenses does not create a lawyer-client or
+other relationship. Creative Commons makes its licenses and related
+information available on an "as-is" basis. Creative Commons gives no
+warranties regarding its licenses, any material licensed under their
+terms and conditions, or any related information. Creative Commons
+disclaims all liability for damages resulting from their use to the
+fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and
+conditions that creators and other rights holders may use to share
+original works of authorship and other material subject to copyright
+and certain other rights specified in the public license below. The
+following considerations are for informational purposes only, are not
+exhaustive, and do not form part of our licenses.
+
+     Considerations for licensors: Our public licenses are
+     intended for use by those authorized to give the public
+     permission to use material in ways otherwise restricted by
+     copyright and certain other rights. Our licenses are
+     irrevocable. Licensors should read and understand the terms
+     and conditions of the license they choose before applying it.
+     Licensors should also secure all rights necessary before
+     applying our licenses so that the public can reuse the
+     material as expected. Licensors should clearly mark any
+     material not subject to the license. This includes other CC-
+     licensed material, or material used under an exception or
+     limitation to copyright. More considerations for licensors:
+    wiki.creativecommons.org/Considerations_for_licensors
+
+     Considerations for the public: By using one of our public
+     licenses, a licensor grants the public permission to use the
+     licensed material under specified terms and conditions. If
+     the licensor's permission is not necessary for any reason--for
+     example, because of any applicable exception or limitation to
+     copyright--then that use is not regulated by the license. Our
+     licenses grant only permissions under copyright and certain
+     other rights that a licensor has authority to grant. Use of
+     the licensed material may still be restricted for other
+     reasons, including because others have copyright or other
+     rights in the material. A licensor may make special requests,
+     such as asking that all changes be marked or described.
+     Although not required by our licenses, you are encouraged to
+     respect those requests where reasonable. More considerations
+     for the public:
+    wiki.creativecommons.org/Considerations_for_licensees
+
+=======================================================================
+
+Creative Commons Attribution-ShareAlike 4.0 International Public
+License
+
+By exercising the Licensed Rights (defined below), You accept and agree
+to be bound by the terms and conditions of this Creative Commons
+Attribution-ShareAlike 4.0 International Public License ("Public
+License"). To the extent this Public License may be interpreted as a
+contract, You are granted the Licensed Rights in consideration of Your
+acceptance of these terms and conditions, and the Licensor grants You
+such rights in consideration of benefits the Licensor receives from
+making the Licensed Material available under these terms and
+conditions.
+
+
+Section 1 -- Definitions.
+
+  a. Adapted Material means material subject to Copyright and Similar
+     Rights that is derived from or based upon the Licensed Material
+     and in which the Licensed Material is translated, altered,
+     arranged, transformed, or otherwise modified in a manner requiring
+     permission under the Copyright and Similar Rights held by the
+     Licensor. For purposes of this Public License, where the Licensed
+     Material is a musical work, performance, or sound recording,
+     Adapted Material is always produced where the Licensed Material is
+     synched in timed relation with a moving image.
+
+  b. Adapter's License means the license You apply to Your Copyright
+     and Similar Rights in Your contributions to Adapted Material in
+     accordance with the terms and conditions of this Public License.
+
+  c. BY-SA Compatible License means a license listed at
+     creativecommons.org/compatiblelicenses, approved by Creative
+     Commons as essentially the equivalent of this Public License.
+
+  d. Copyright and Similar Rights means copyright and/or similar rights
+     closely related to copyright including, without limitation,
+     performance, broadcast, sound recording, and Sui Generis Database
+     Rights, without regard to how the rights are labeled or
+     categorized. For purposes of this Public License, the rights
+     specified in Section 2(b)(1)-(2) are not Copyright and Similar
+     Rights.
+
+  e. Effective Technological Measures means those measures that, in the
+     absence of proper authority, may not be circumvented under laws
+     fulfilling obligations under Article 11 of the WIPO Copyright
+     Treaty adopted on December 20, 1996, and/or similar international
+     agreements.
+
+  f. Exceptions and Limitations means fair use, fair dealing, and/or
+     any other exception or limitation to Copyright and Similar Rights
+     that applies to Your use of the Licensed Material.
+
+  g. License Elements means the license attributes listed in the name
+     of a Creative Commons Public License. The License Elements of this
+     Public License are Attribution and ShareAlike.
+
+  h. Licensed Material means the artistic or literary work, database,
+     or other material to which the Licensor applied this Public
+     License.
+
+  i. Licensed Rights means the rights granted to You subject to the
+     terms and conditions of this Public License, which are limited to
+     all Copyright and Similar Rights that apply to Your use of the
+     Licensed Material and that the Licensor has authority to license.
+
+  j. Licensor means the individual(s) or entity(ies) granting rights
+     under this Public License.
+
+  k. Share means to provide material to the public by any means or
+     process that requires permission under the Licensed Rights, such
+     as reproduction, public display, public performance, distribution,
+     dissemination, communication, or importation, and to make material
+     available to the public including in ways that members of the
+     public may access the material from a place and at a time
+     individually chosen by them.
+
+  l. Sui Generis Database Rights means rights other than copyright
+     resulting from Directive 96/9/EC of the European Parliament and of
+     the Council of 11 March 1996 on the legal protection of databases,
+     as amended and/or succeeded, as well as other essentially
+     equivalent rights anywhere in the world.
+
+  m. You means the individual or entity exercising the Licensed Rights
+     under this Public License. Your has a corresponding meaning.
+
+
+Section 2 -- Scope.
+
+  a. License grant.
+
+       1. Subject to the terms and conditions of this Public License,
+          the Licensor hereby grants You a worldwide, royalty-free,
+          non-sublicensable, non-exclusive, irrevocable license to
+          exercise the Licensed Rights in the Licensed Material to:
+
+            a. reproduce and Share the Licensed Material, in whole or
+               in part; and
+
+            b. produce, reproduce, and Share Adapted Material.
+
+       2. Exceptions and Limitations. For the avoidance of doubt, where
+          Exceptions and Limitations apply to Your use, this Public
+          License does not apply, and You do not need to comply with
+          its terms and conditions.
+
+       3. Term. The term of this Public License is specified in Section
+          6(a).
+
+       4. Media and formats; technical modifications allowed. The
+          Licensor authorizes You to exercise the Licensed Rights in
+          all media and formats whether now known or hereafter created,
+          and to make technical modifications necessary to do so. The
+          Licensor waives and/or agrees not to assert any right or
+          authority to forbid You from making technical modifications
+          necessary to exercise the Licensed Rights, including
+          technical modifications necessary to circumvent Effective
+          Technological Measures. For purposes of this Public License,
+          simply making modifications authorized by this Section 2(a)
+          (4) never produces Adapted Material.
+
+       5. Downstream recipients.
+
+            a. Offer from the Licensor -- Licensed Material. Every
+               recipient of the Licensed Material automatically
+               receives an offer from the Licensor to exercise the
+               Licensed Rights under the terms and conditions of this
+               Public License.
+
+            b. Additional offer from the Licensor -- Adapted Material.
+               Every recipient of Adapted Material from You
+               automatically receives an offer from the Licensor to
+               exercise the Licensed Rights in the Adapted Material
+               under the conditions of the Adapter's License You apply.
+
+            c. No downstream restrictions. You may not offer or impose
+               any additional or different terms or conditions on, or
+               apply any Effective Technological Measures to, the
+               Licensed Material if doing so restricts exercise of the
+               Licensed Rights by any recipient of the Licensed
+               Material.
+
+       6. No endorsement. Nothing in this Public License constitutes or
+          may be construed as permission to assert or imply that You
+          are, or that Your use of the Licensed Material is, connected
+          with, or sponsored, endorsed, or granted official status by,
+          the Licensor or others designated to receive attribution as
+          provided in Section 3(a)(1)(A)(i).
+
+  b. Other rights.
+
+       1. Moral rights, such as the right of integrity, are not
+          licensed under this Public License, nor are publicity,
+          privacy, and/or other similar personality rights; however, to
+          the extent possible, the Licensor waives and/or agrees not to
+          assert any such rights held by the Licensor to the limited
+          extent necessary to allow You to exercise the Licensed
+          Rights, but not otherwise.
+
+       2. Patent and trademark rights are not licensed under this
+          Public License.
+
+       3. To the extent possible, the Licensor waives any right to
+          collect royalties from You for the exercise of the Licensed
+          Rights, whether directly or through a collecting society
+          under any voluntary or waivable statutory or compulsory
+          licensing scheme. In all other cases the Licensor expressly
+          reserves any right to collect such royalties.
+
+
+Section 3 -- License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the
+following conditions.
+
+  a. Attribution.
+
+       1. If You Share the Licensed Material (including in modified
+          form), You must:
+
+            a. retain the following if it is supplied by the Licensor
+               with the Licensed Material:
+
+                 i. identification of the creator(s) of the Licensed
+                    Material and any others designated to receive
+                    attribution, in any reasonable manner requested by
+                    the Licensor (including by pseudonym if
+                    designated);
+
+                ii. a copyright notice;
+
+               iii. a notice that refers to this Public License;
+
+                iv. a notice that refers to the disclaimer of
+                    warranties;
+
+                 v. a URI or hyperlink to the Licensed Material to the
+                    extent reasonably practicable;
+
+            b. indicate if You modified the Licensed Material and
+               retain an indication of any previous modifications; and
+
+            c. indicate the Licensed Material is licensed under this
+               Public License, and include the text of, or the URI or
+               hyperlink to, this Public License.
+
+       2. You may satisfy the conditions in Section 3(a)(1) in any
+          reasonable manner based on the medium, means, and context in
+          which You Share the Licensed Material. For example, it may be
+          reasonable to satisfy the conditions by providing a URI or
+          hyperlink to a resource that includes the required
+          information.
+
+       3. If requested by the Licensor, You must remove any of the
+          information required by Section 3(a)(1)(A) to the extent
+          reasonably practicable.
+
+  b. ShareAlike.
+
+     In addition to the conditions in Section 3(a), if You Share
+     Adapted Material You produce, the following conditions also apply.
+
+       1. The Adapter's License You apply must be a Creative Commons
+          license with the same License Elements, this version or
+          later, or a BY-SA Compatible License.
+
+       2. You must include the text of, or the URI or hyperlink to, the
+          Adapter's License You apply. You may satisfy this condition
+          in any reasonable manner based on the medium, means, and
+          context in which You Share Adapted Material.
+
+       3. You may not offer or impose any additional or different terms
+          or conditions on, or apply any Effective Technological
+          Measures to, Adapted Material that restrict exercise of the
+          rights granted under the Adapter's License You apply.
+
+
+Section 4 -- Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that
+apply to Your use of the Licensed Material:
+
+  a. for the avoidance of doubt, Section 2(a)(1) grants You the right
+     to extract, reuse, reproduce, and Share all or a substantial
+     portion of the contents of the database;
+
+  b. if You include all or a substantial portion of the database
+     contents in a database in which You have Sui Generis Database
+     Rights, then the database in which You have Sui Generis Database
+     Rights (but not its individual contents) is Adapted Material,
+     including for purposes of Section 3(b); and
+
+  c. You must comply with the conditions in Section 3(a) if You Share
+     all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not
+replace Your obligations under this Public License where the Licensed
+Rights include other Copyright and Similar Rights.
+
+
+Section 5 -- Disclaimer of Warranties and Limitation of Liability.
+
+  a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
+     EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
+     AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
+     ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
+     IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
+     WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
+     PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
+     ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
+     KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
+     ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
+
+  b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
+     TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
+     NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
+     INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
+     COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
+     USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
+     ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
+     DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
+     IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
+
+  c. The disclaimer of warranties and limitation of liability provided
+     above shall be interpreted in a manner that, to the extent
+     possible, most closely approximates an absolute disclaimer and
+     waiver of all liability.
+
+
+Section 6 -- Term and Termination.
+
+  a. This Public License applies for the term of the Copyright and
+     Similar Rights licensed here. However, if You fail to comply with
+     this Public License, then Your rights under this Public License
+     terminate automatically.
+
+  b. Where Your right to use the Licensed Material has terminated under
+     Section 6(a), it reinstates:
+
+       1. automatically as of the date the violation is cured, provided
+          it is cured within 30 days of Your discovery of the
+          violation; or
+
+       2. upon express reinstatement by the Licensor.
+
+     For the avoidance of doubt, this Section 6(b) does not affect any
+     right the Licensor may have to seek remedies for Your violations
+     of this Public License.
+
+  c. For the avoidance of doubt, the Licensor may also offer the
+     Licensed Material under separate terms or conditions or stop
+     distributing the Licensed Material at any time; however, doing so
+     will not terminate this Public License.
+
+  d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
+     License.
+
+
+Section 7 -- Other Terms and Conditions.
+
+  a. The Licensor shall not be bound by any additional or different
+     terms or conditions communicated by You unless expressly agreed.
+
+  b. Any arrangements, understandings, or agreements regarding the
+     Licensed Material not stated herein are separate from and
+     independent of the terms and conditions of this Public License.
+
+
+Section 8 -- Interpretation.
+
+  a. For the avoidance of doubt, this Public License does not, and
+     shall not be interpreted to, reduce, limit, restrict, or impose
+     conditions on any use of the Licensed Material that could lawfully
+     be made without permission under this Public License.
+
+  b. To the extent possible, if any provision of this Public License is
+     deemed unenforceable, it shall be automatically reformed to the
+     minimum extent necessary to make it enforceable. If the provision
+     cannot be reformed, it shall be severed from this Public License
+     without affecting the enforceability of the remaining terms and
+     conditions.
+
+  c. No term or condition of this Public License will be waived and no
+     failure to comply consented to unless expressly agreed to by the
+     Licensor.
+
+  d. Nothing in this Public License constitutes or may be interpreted
+     as a limitation upon, or waiver of, any privileges and immunities
+     that apply to the Licensor or You, including from the legal
+     processes of any jurisdiction or authority.
+
+
+=======================================================================
+
+Creative Commons is not a party to its public
+licenses. Notwithstanding, Creative Commons may elect to apply one of
+its public licenses to material it publishes and in those instances
+will be considered the “Licensor.” The text of the Creative Commons
+public licenses is dedicated to the public domain under the CC0 Public
+Domain Dedication. Except for the limited purpose of indicating that
+material is shared under a Creative Commons public license or as
+otherwise permitted by the Creative Commons policies published at
+creativecommons.org/policies, Creative Commons does not authorize the
+use of the trademark "Creative Commons" or any other trademark or logo
+of Creative Commons without its prior written consent including,
+without limitation, in connection with any unauthorized modifications
+to any of its public licenses or any other arrangements,
+understandings, or agreements concerning use of licensed material. For
+the avoidance of doubt, this paragraph does not form part of the
+public licenses.
+
+Creative Commons may be contacted at creativecommons.org.
+

README.md

@@ -0,0 +1,96 @@
+---
+language: pl
+license: cc-by-sa-4.0
+library_name: matcha-tts
+pipeline_tag: text-to-speech
+base_model: machinekind/Matcha-TTS-PL
+tags: [tts, polish, matcha-tts, flow-matching, hifigan, distillation, voxcpm2, code-switching, robot, onnx]
+---
+
+# Matcha-TTS-PL v2 — one synthetic voice, modern Polish, English words
+
+Second version of [Matcha-TTS-PL](https://huggingface.co/machinekind/Matcha-TTS-PL): the same small, fast,
+non-autoregressive Matcha-TTS (20.9 M parameters, real-time on a small GPU or in a browser) **distilled from VoxCPM2**,
+a 2B-parameter TTS, onto a single synthetic voice. What changes compared with v1:
+
+- **One original voice, `fav2`** — designed from a text description with VoxCPM2 (OpenBMB, Apache-2.0) and frozen as a
+  prompt cache; not a recording or a clone of a real person (closest of our reference readers: speaker similarity 0.83).
+- **Modern Polish of a humanoid assistant** — 13.1 h of training speech: assistant speech in 18 everyday domains, modern
+  corpora (NKJP, Wikinews, KPWr, ParlaMint, OpenAssistant, Wikibooks, Wikivoyage) and reviewed YouTube transcripts;
+  two thirds of it in multi-sentence passages (10–25 s), so prosody is planned across sentences.
+- **English words inside Polish sentences** — "Po meetingu wyślę ci feedback na maila" is read with English phonemes:
+  English words are detected automatically and phonemized by espeak-ng en-us, the rest by espeak-ng pl
+  (`english/`). No pronunciation dictionary needed for common loanwords and brands.
+- **A HiFi-GAN fine-tuned on this voice**, with real ground truth (the teacher's audio) — the v1 vocoder was tuned for the
+  audiobook readers and blurs this voice.
+
+Try it in the browser: [spaces/machinekind/matcha-tts-pl-v2](https://huggingface.co/spaces/machinekind/matcha-tts-pl-v2).
+
+## Quality (12 held-out sentences, Mac CPU, 4 ODE steps, temperature 0.8)
+
+| system | UTMOS ↑ | Whisper WER ↓ | similarity to fav2 ↑ |
+|---|---|---|---|
+| VoxCPM2 teacher (2B, autoregressive) | 3.68 | 1.3 % | 0.883 |
+| **v2** (this model + its vocoder) | 3.40 | 1.3 % | 0.850 |
+| v1 model with the new voice slot, before fine-tuning | 2.91 | 2.7 % | 0.725 |
+
+WER counts Whisper's spelling too: the teacher's two "errors" are "krutki" for *krótki* and "HWSR" for the robot's name.
+Speaking rate: the released default **tempo is 0.8** (`length_scale`); at 1.0 v2 speaks about 19 % slower than the teacher.
+
+## Files
+
+| path | what |
+|---|---|
+| `onnx/matcha_pl_t4.onnx`, `onnx/matcha_pl_t2.onnx` | acoustic model + fav2 vocoder in one graph, 4 / 2 ODE steps; inputs `x` (phoneme ids), `x_lengths`, `scales=[temperature, length_scale]`, `spk_emb` (float32 [1, 64]); outputs `wav` (22.05 kHz), `wav_lengths` |
+| `onnx/voices.json` | the fav2 speaker embedding, 18 style embeddings, defaults |
+| `model/matcha_pl_v2.ckpt` | PyTorch checkpoint (Matcha-TTS + the tts-pl patch: style tokens, Polish cleaners) — speaker 20 = fav2 |
+| `vocoder/generator_fav2`, `vocoder/config.json` | HiFi-GAN (v1 architecture) fine-tuned on this model's mels ↔ the fav2 audio |
+| `english/` | English detection (`english_detect.py`, `english_words.txt`; browser: `english.js`, `english_data.json`) and the mixed phonemizer (`polish_mixed_cleaners.py`; browser: `cleaner.js`) |
+| `samples/` | the 12 held-out sentences, v2 and teacher |
+| `data/` | style token map, speaker map |
+| `ATTRIBUTION.md` | every text source of the training data, and the base model's data |
+
+## Quick start (ONNX, Python)
+
+```python
+import sys, json, numpy as np, onnxruntime as ort, soundfile as sf
+sys.path.append("english")                          # english_detect.py + english_words.txt (optional fallback: pip install wordfreq)
+from english_detect import mark_english
+from matcha.text import text_to_sequence            # Matcha-TTS with the tts-pl cleaners: paste english/polish_cleaners.py and
+from matcha.utils.utils import intersperse          # english/polish_mixed_cleaners.py into matcha/text/cleaners.py
+text = mark_english("Po meetingu wyślę ci feedback na maila, a w weekend zrobię update.")
+ids = np.array(intersperse(text_to_sequence(text, ["polish_mixed_cleaners"])[0], 0), np.int64)
+vj = json.load(open("onnx/voices.json"))
+spk = np.array(vj["speakers"]["0"]["emb"], np.float32) + np.array(vj["styles"]["7"]["emb"], np.float32)   # 7 = neutral style
+sess = ort.InferenceSession("onnx/matcha_pl_t4.onnx")
+wav, n = sess.run(None, {"x": ids[None], "x_lengths": np.array([ids.size]), "scales": np.array([0.8, 0.8], np.float32), "spk_emb": spk[None]})
+sf.write("out.wav", wav[0, :n[0]], 22050)   # scales = [temperature, length_scale (tempo)]
+```
+
+Text front end: `mark_english()` wraps English words in braces (`Po {meeting}u wyślę ci {feedback}.`), the cleaner
+phonemizes braced spans with espeak-ng `en-us` and the rest with `pl` (with the Polish ending of an inflected English
+word glued on unstressed, and one-letter prepositions before English words handled by rule). You can brace any word
+yourself. The browser port produces the training phonemes character for character (tested on 300 sentences).
+
+## Training
+
+- Base: Matcha-TTS-PL v1 (target model), new speaker row initialised from the closest real reader; same mel statistics,
+  style tokens and phoneme table (all en-us phones were already in the symbol table, now trained).
+- Data: 6 493 clips, 13.1 h, generated by VoxCPM2 with the frozen fav2 voice (anchor + continuation, 10 diffusion
+  steps), every clip gated by Whisper (text match), UTMOS ≥ 3.2 and speaker similarity ≥ 0.80, up to 3 takes per
+  sentence. Source texts reviewed and minimally completed by Claude (cut sentences finished, numbers written as words).
+- Matcha fine-tune: 10 k steps, batch 64, lr 5e-5, bf16, RTX 6000 Ada (≈ 1.5 h); quality plateaued from 5 k steps.
+- Vocoder: teacher-forced mels of the 10 k checkpoint for all training clips paired with the VoxCPM2 audio, HiFi-GAN
+  universal v1 fine-tuned 30 k steps, lr 2e-5.
+
+## Limitations
+
+- One voice. The rhythm is more regular than the teacher's: Matcha's deterministic duration predictor averages timing.
+- English detection uses a curated list (≈ 1 900 words and brands, Polish lookalikes excluded); an unlisted English
+  word is read the Polish way unless you put it in braces.
+- Synthetic voice: label generated speech as synthetic; don't present it as a real person.
+
+## Licence
+
+CC BY-SA 4.0 (training data includes CC BY-SA texts; the base model is CC BY-SA 4.0). VoxCPM2 is Apache-2.0;
+espeak-ng (runtime phonemizer) is GPL-3.0. See `ATTRIBUTION.md`.

data/speaker_map.json

@@ -0,0 +1,86 @@
+{
+ "0": {
+  "source": "data/wl",
+  "speaker": "Bartosz Bielenia"
+ },
+ "1": {
+  "source": "data/wl",
+  "speaker": "Katarzyna Faszczewska"
+ },
+ "2": {
+  "source": "data/wl",
+  "speaker": "Wojciech Masiak"
+ },
+ "3": {
+  "source": "data/wl",
+  "speaker": "Bartosz Głogowski"
+ },
+ "4": {
+  "source": "data/wl",
+  "speaker": "Jan Staszczyk"
+ },
+ "5": {
+  "source": "data/wl",
+  "speaker": "Marek Proszek"
+ },
+ "6": {
+  "source": "data/wl",
+  "speaker": "Piotr Kopa"
+ },
+ "7": {
+  "source": "data/wl",
+  "speaker": "Radosław Krzyżowski"
+ },
+ "8": {
+  "source": "data/wl",
+  "speaker": "Andrzej Seweryn"
+ },
+ "9": {
+  "source": "data/wl",
+  "speaker": "Magdalena Gaweł"
+ },
+ "10": {
+  "source": "data/wl",
+  "speaker": "Jakub Sasak"
+ },
+ "11": {
+  "source": "data/wl",
+  "speaker": "Ewelina Pankowska"
+ },
+ "12": {
+  "source": "data/wl",
+  "speaker": "Krzysztof Zarzecki"
+ },
+ "13": {
+  "source": "data/wl",
+  "speaker": "Joanna Domańska"
+ },
+ "14": {
+  "source": "data/wl",
+  "speaker": "Bartosz Mazur"
+ },
+ "15": {
+  "source": "data/azon",
+  "speaker": "azon_spont_102903M"
+ },
+ "16": {
+  "source": "data/azon",
+  "speaker": "azon_spont_99321M"
+ },
+ "17": {
+  "source": "data/azon",
+  "speaker": "azon_spont_99322K"
+ },
+ "18": {
+  "source": "data/azon",
+  "speaker": "azon_spont_99324K"
+ },
+ "19": {
+  "source": "data/azon",
+  "speaker": "azon_spont_99329M"
+ },
+ "20": {
+  "source": "VoxCPM2 (synthetic)",
+  "speaker": "fav2"
+ }
+}

data/style_map.json

@@ -0,0 +1,111 @@
+{
+ "0": {
+  "pitch_range": "flat",
+  "rate": "slow",
+  "question": false,
+  "label": "flat-range · slow"
+ },
+ "1": {
+  "pitch_range": "flat",
+  "rate": "slow",
+  "question": true,
+  "label": "flat-range · slow · question"
+ },
+ "2": {
+  "pitch_range": "flat",
+  "rate": "normal",
+  "question": false,
+  "label": "flat-range · normal"
+ },
+ "3": {
+  "pitch_range": "flat",
+  "rate": "normal",
+  "question": true,
+  "label": "flat-range · normal · question"
+ },
+ "4": {
+  "pitch_range": "flat",
+  "rate": "fast",
+  "question": false,
+  "label": "flat-range · fast"
+ },
+ "5": {
+  "pitch_range": "flat",
+  "rate": "fast",
+  "question": true,
+  "label": "flat-range · fast · question"
+ },
+ "6": {
+  "pitch_range": "mid",
+  "rate": "slow",
+  "question": false,
+  "label": "mid-range · slow"
+ },
+ "7": {
+  "pitch_range": "mid",
+  "rate": "slow",
+  "question": true,
+  "label": "mid-range · slow · question"
+ },
+ "8": {
+  "pitch_range": "mid",
+  "rate": "normal",
+  "question": false,
+  "label": "mid-range · normal"
+ },
+ "9": {
+  "pitch_range": "mid",
+  "rate": "normal",
+  "question": true,
+  "label": "mid-range · normal · question"
+ },
+ "10": {
+  "pitch_range": "mid",
+  "rate": "fast",
+  "question": false,
+  "label": "mid-range · fast"
+ },
+ "11": {
+  "pitch_range": "mid",
+  "rate": "fast",
+  "question": true,
+  "label": "mid-range · fast · question"
+ },
+ "12": {
+  "pitch_range": "wide",
+  "rate": "slow",
+  "question": false,
+  "label": "wide-range · slow"
+ },
+ "13": {
+  "pitch_range": "wide",
+  "rate": "slow",
+  "question": true,
+  "label": "wide-range · slow · question"
+ },
+ "14": {
+  "pitch_range": "wide",
+  "rate": "normal",
+  "question": false,
+  "label": "wide-range · normal"
+ },
+ "15": {
+  "pitch_range": "wide",
+  "rate": "normal",
+  "question": true,
+  "label": "wide-range · normal · question"
+ },
+ "16": {
+  "pitch_range": "wide",
+  "rate": "fast",
+  "question": false,
+  "label": "wide-range · fast"
+ },
+ "17": {
+  "pitch_range": "wide",
+  "rate": "fast",
+  "question": true,
+  "label": "wide-range · fast · question"
+ },
+ "neutral": "7"
+}

english/cleaner.js

@@ -0,0 +1,102 @@
+// Browser port of matcha_patch/polish_cleaners.py + polish_mixed_cleaners: Polish text normalisation, espeak-ng (WASM) IPA,
+// English words in {braces} phonemized with the en-us voice, symbol ids. Must match the Python cleaners character for character.
+import { initialize, setVoice, getPhonemes } from "https://cdn.jsdelivr.net/npm/espeak-phonemizer@0.1.2/src/espeak.mjs";
+import { SYMBOL_TO_ID } from "./symbols.js";
+
+const PL_WORD = "0-9a-ząćęłńóśźż";
+const esc = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+function abbr(k, v) {
+  if (k === "r." || k === "w.") return [new RegExp(`(?<=[0-9ivx])\\s?${esc(k)}`, "g"), ` ${v}`];
+  const tail = k.endsWith(".") ? "" : `(?![${PL_WORD}])`;
+  return [new RegExp(`(?<![${PL_WORD}])${esc(k)}${tail}`, "g"), v];
+}
+const ABBREV = [["np.", "na przykład"], ["tzn.", "to znaczy"], ["tj.", "to jest"], ["itd.", "i tak dalej"], ["itp.", "i tym podobne"],
+  ["m.in.", "między innymi"], ["dr", "doktor"], ["prof.", "profesor"], ["inż.", "inżynier"], ["mgr", "magister"], ["ul.", "ulica"],
+  ["al.", "aleja"], ["r.", "roku"], ["w.", "wieku"], ["godz.", "godzina"], ["min.", "minut"], ["św.", "święty"], ["nr", "numer"],
+  ["tel.", "telefon"], ["zł", "złotych"]].map(([k, v]) => abbr(k, v));
+
+export function normalizeText(text) {
+  let t = text.toLowerCase();
+  t = t.replace(/\s*(?:—|–| - )\s*/g, ", ").replace(/…/g, "...");
+  t = t.replace(/[„”“"«»]/g, "");
+  for (const [rx, rep] of ABBREV) t = t.replace(rx, rep);
+  return t;
+}
+
+let ready = null, voice = "";
+export function initPhonemizer() {
+  if (!ready) ready = (async () => { await initialize("https://cdn.jsdelivr.net/npm/espeak-phonemizer@0.1.2/dist"); await setVoice("pl"); voice = "pl"; })();
+  return ready;
+}
+async function useVoice(v) { if (voice !== v) { await setVoice(v); voice = v; } }
+
+// phonemizer (python) splits on punctuation, phonemizes each chunk and glues the punctuation back
+const PUNCT = /([;:,.!?]+)/;
+function espeakChunks(t) {
+  let out = "";
+  for (const part of t.split(PUNCT)) {
+    if (!part) continue;
+    if (/^[;:,.!?]+$/.test(part)) { out += part; continue; }
+    const chunk = part.trim();
+    if (!chunk) continue;
+    const ph = getPhonemes(chunk).map(c => (c.phonemes || "").trim()).filter(Boolean).join(" ");
+    if (!ph) continue;
+    out += (out && !out.endsWith(" ") ? " " : "") + ph;
+  }
+  return out.replace(/\s+/g, " ").trim();
+}
+// this espeak-ng WASM build writes "iʲ" before a vowel ("znajomymiʲ albo"); the training phonemizer (espeak-ng 1.50, Linux)
+// never does ("iʲ" does not occur in the training data at all) -> normalise, so the model gets what it was trained on
+const fixPl = (ph) => ph.replace(/iʲ/g, "i");
+async function plPhon(text) { await useVoice("pl"); const t = normalizeText(text); return t.trim() ? fixPl(espeakChunks(t)) : ""; }
+async function enPhon(text) { await useVoice("en-us"); return espeakChunks(text).replace(/ʲ/g, ""); }   // same WASM quirk; en-us never has ʲ
+
+const VOICELESS = ["p", "t", "k", "f", "s", "ʃ", "θ", "h", "tʃ", "ˈp", "ˈt", "ˈk", "ˈf", "ˈs", "ˈʃ", "ˈθ", "ˈh", "ˌp", "ˌt", "ˌk", "ˌf", "ˌs", "ˌh"];
+function oneLetter(w, nextPh) {
+  const vl = VOICELESS.some(p => nextPh.startsWith(p));
+  return { w: vl ? "f" : "v", z: vl ? "s" : "z", i: "i", a: "a", o: "ɔ", u: "u" }[w];
+}
+const unstress = (ph) => { const i = ph.indexOf(" "); const head = i < 0 ? ph : ph.slice(0, i); return head.replace(/[ˈˌ]/g, "") + (i < 0 ? "" : ph.slice(i)); };
+
+export async function textToIpa(text) {
+  await initPhonemizer();
+  if (!text.includes("{")) return plPhon(text);
+  let out = "", pos = 0;
+  for (const m of text.matchAll(/\{([^{}]+)\}/g)) {
+    let pl = text.slice(pos, m.index), prep = "";
+    const mm = pl.match(/(?:^|\s)([wzaiouWZAIOU])\s*$/);
+    if (mm) { prep = mm[1].toLowerCase(); pl = pl.slice(0, mm.index + mm[0].indexOf(mm[1])); }   // = Python pl[:mm.start(1)]
+    if (pl.trim()) {
+      const glue = !/^\s/.test(pl) && out !== "";
+      let ph = await plPhon(pl);
+      if (glue) ph = unstress(ph);
+      out += (glue ? "" : " ") + ph;
+    }
+    const en = await enPhon(m[1]);
+    if (prep) out += " " + oneLetter(prep, en);
+    const glue = m.index > 0 && !/\s/.test(text[m.index - 1]) && out !== "";
+    out += (glue ? "" : " ") + en;
+    pos = m.index + m[0].length;
+  }
+  const tail = text.slice(pos);
+  if (tail.trim()) {
+    const glue = !/^\s/.test(tail);
+    let ph = await plPhon(tail);
+    if (glue) ph = unstress(ph);
+    out += (glue ? "" : " ") + ph;
+  }
+  await useVoice("pl");
+  return out.replace(/\s+/g, " ").trim();
+}
+
+export function ipaToIds(ipa) {
+  const ids = [];
+  const unknown = new Set();
+  for (const ch of ipa) {
+    const id = SYMBOL_TO_ID[ch];
+    if (id === undefined) unknown.add(ch); else ids.push(id);
+  }
+  const seq = new Array(ids.length * 2 + 1).fill(0);
+  ids.forEach((v, i) => { seq[2 * i + 1] = v; });
+  return { seq, unknown: [...unknown] };
+}

english/english.js

@@ -0,0 +1,32 @@
+// Browser port of voxcpm/english_detect.py: wrap English words of a Polish sentence in braces ("Po {meeting}u wyślę ci {feedback}.")
+// for the mixed Polish/English phonemizer. Curated list only (the Python version also has a word-frequency fallback for
+// unlisted words; a frequency table is too big for the browser). Inflected forms whose full form is a common Polish word
+// (wina, legendy, deskę) are precomputed in english_data.json -> pl_inflected_block.
+let D = null;
+export async function initEnglish(url = new URL("./english_data.json", import.meta.url)) {
+  if (!D) {
+    const j = await (await fetch(url)).json();
+    D = { cur: new Set(j.curated), rap: new Set(j.read_as_polish), suf: j.suffixes, loan: j.loan_stems, block: new Set(j.pl_inflected_block) };
+  }
+  return D;
+}
+export function setEnglishData(j) { D = { cur: new Set(j.curated), rap: new Set(j.read_as_polish), suf: j.suffixes, loan: j.loan_stems, block: new Set(j.pl_inflected_block) }; }
+const known = (w) => { const l = w.toLowerCase(); return D.cur.has(l) && !D.rap.has(l); };
+function splitWord(w) {
+  if (known(w)) return [w, ""];
+  const lw = w.toLowerCase();
+  if (w.includes("'") || w.includes("’")) {
+    const st = w.replace(/’/g, "'").split("'");
+    const head = st.shift(), rest = st.join("'");
+    if (known(head)) return [head, rest ? "'" + rest : ""];
+  }
+  if (D.block.has(lw)) return null;
+  for (const k of D.suf) if (lw.endsWith(k) && w.length - k.length >= 3 && known(w.slice(0, -k.length))) return [w.slice(0, -k.length), w.slice(-k.length)];
+  return null;
+}
+const WORD = /\{[^{}]*\}|[A-Za-zÀ-ÿĄĆĘŁŃÓŚŹŻąćęłńóśźż][A-Za-zÀ-ÿĄĆĘŁŃÓŚŹŻąćęłńóśźż'’-]*/g;
+export function markEnglish(text) {
+  if (!D) throw new Error("initEnglish() first");
+  const out = text.replace(WORD, (w) => { if (w.startsWith("{")) return w; const s = splitWord(w); return s ? `{${s[0]}}${s[1]}` : w; });
+  return out.replace(/\}(\s+)\{/g, "$1");
+}

english/english_data.json

@@ -0,0 +1 @@
+{"curated": ["ac/dc", "accelerator", "access", "account", "account manager", "accountability", "achievement", "acquisition", "action", "action points", "add-on", "adobe", "adopter", "ads", "aesthetic", "affiliate", "after", "after party", "afterparty", "afterwork", "agile", "air", "airbnb", "airbus", "airdrop", "airplay", "airpods", "airtag", "alert", "algorithm", "aliexpress", "alignment", "all", "all inclusive", "all-star", "alternative", "amazing", "amazon", "amazon prime", "ambassador", "analyst", "analytics", "and", "angel", "angular", "anyway", "app", "app store", "apple", "apple music", "apple pay", "apple store", "apple tv", "apple watch", "apps", "art", "art director", "asap", "attachment", "audiobook", "augmented", "augmented reality", "automation", "autoplay", "awareness", "away", "awesome", "azure", "babe", "baby", "baby boomers", "baby shower", "babysitter", "back", "back office", "back-end", "backdoor", "backend", "background", "backlog", "backpack", "backpacker", "backpacking", "backstage", "backup", "bacon", "badge", "bae", "balance", "balancer", "bank", "barbecue", "barber", "barbershop", "barbie", "bartender", "baseball", "basic", "basically", "basketball", "batman", "battle", "battle royale", "beatbox", "beauty", "beef", "behind", "behind the scenes", "benchmark", "benchmarking", "benefits", "best", "best friend", "best practice", "best practices", "best seller", "bestie", "big", "big brother", "big data", "billboard", "binge", "binge watching", "binge-watching", "bio", "bird", "birthday", "bitcoin", "black", "black friday", "blame", "blind", "blind date", "bling", "blockbuster", "blockchain", "blocker", "blogger", "bluetooth", "boarding", "body", "body positive", "body shaming", "bodybuilder", "bodybuilding", "boeing", "boilerplate", "bolt", "bombing", "bond", "booking", "boomer", "boomers", "booster", "bootcamp", "boss", "boss fight", "bottleneck", "bounty", "bowl", "bowling", "box", "box office", "boyfriend", "boysband", "brainstorm", "brainstorming", "branch", "brand", "brand ambassador", "brand awareness", "brandbook", "branding", "break", "breakdance", "breakdown", "breakfast", "breaking", "breaking news", "breakpoint", "brexit", "brief", "briefing", "brother", "brownie", "brunch", "bubble", "bubble tea", "bucket", "bucket list", "budget", "bug", "bug bounty", "bug report", "bugfix", "bugs", "build", "building", "bull", "bumble", "bundle", "burger", "burger king", "burnout", "burpees", "business", "business angel", "business case", "business intelligence", "business model", "business partner", "business plan", "button", "buzz", "buzzword", "by the way", "bye", "c-level", "cache", "calisthenics", "call", "call to action", "cancel", "cancel culture", "capcut", "capital", "cappuccino", "caption", "car sharing", "card", "cardio", "career", "career path", "carpet", "carry", "carsharing", "case", "case study", "cash", "cash flow", "cashback", "casual", "casual friday", "catcalling", "catch-up", "catering", "challenge", "champions", "champions league", "change", "changelog", "changer", "chargeback", "charger", "chat", "chat gpt", "chatbot", "chatgpt", "cheat", "cheat sheet", "check-in", "check-out", "checkbox", "checking", "checklist", "checkout", "checkpoint", "cheers", "cheeseburger", "cheesecake", "chill", "chillout", "christmas", "chromebook", "churn", "churn rate", "cider", "city", "city break", "claim", "clean", "clean code", "clearance", "click", "clickbait", "clickup", "cliffhanger", "climate", "climate change", "clone", "close", "close friends", "closing", "cloud", "cloud computing", "cloudflare", "cluster", "clutch", "co-host", "co-op", "coach", "coachella", "coaching", "code", "code review", "codebase", "coder", "coding", "coffee", "coffee shop", "cold", "cold call", "cold mail", "coldplay", "coliving", "collab", "collabs", "comeback", "comedy", "coming", "coming out", "coming-of-age", "comment", "commit", "commitment", "community", "community manager", "compiler", "compliance", "component", "computer", "computer vision", "computing", "concealer", "concept", "conference", "conference call", "confluence", "congrats", "congratulations", "consent", "contactless", "container", "content", "content creator", "content marketing", "continue", "continuous", "continuous integration", "contract", "converse", "conversion", "conversion rate", "cookie", "cookies", "cool", "copilot", "copy", "copy paste", "copy-paste", "copyright", "copywriter", "copywriting", "corporate", "cosmopolitan", "cosplay", "counter-strike", "country", "coupon", "court", "cover", "coworking", "cozy", "craft", "creative", "creative director", "creator", "credit", "credit card", "creep", "creepy", "crew", "crime", "cringe", "cronjob", "cross-platform", "cross-sell", "crossfit", "crossover", "crowdfunding", "crunch", "crush", "crypto", "culture", "cupcake", "curling", "customer", "customer experience", "customer service", "customer success", "customer support", "cut-off", "cute", "cutscene", "cyber", "cyber monday", "cyberbullying", "cyberpunk", "daily", "daily standup", "dark", "dark mode", "darling", "darts", "dashboard", "data analyst", "data engineer", "data science", "data scientist", "dataset", "date", "dating", "deadlift", "deadline", "deadlock", "deadpool", "deal", "dealer", "debriefing", "debt", "debug", "debugger", "debugging", "deck", "deep", "deep dive", "deep learning", "deep work", "deepfake", "deepseek", "deliverable", "delivery", "dell", "dependency", "deploy", "deployment", "design", "design patterns", "designer", "desktop", "detox", "dev", "developer", "devops", "devtools", "diff", "digital", "digital detox", "digital nomad", "dinner", "dip", "direct", "direct message", "director", "discord", "discount", "discovery", "disney", "disneyland", "diss", "dive", "diversity", "dj", "docker", "docs", "dog", "done", "donut", "double", "double click", "downgrade", "downtime", "draft", "drag", "drag and drop", "dream", "dream job", "dream team", "dress", "dress code", "drip", "drive", "drive-in", "drive-thru", "drop", "dropbox", "dropdown", "dropshipping", "dubstep", "duty-free", "déjà", "déjà vu", "e-book", "e-commerce", "e-learning", "e-mail", "earbuds", "early", "early access", "early adopter", "early bird", "easter", "easter egg", "easy", "easyjet", "eats", "ebay", "ebook", "eco-friendly", "ecommerce", "edge", "edge case", "egg", "elasticsearch", "elearning", "elevator", "elevator pitch", "email", "embedding", "eminem", "emirates", "emmy", "emoji", "emote", "employer", "employer branding", "empowerment", "encryption", "end-to-end", "endpoint", "engagement", "engine", "engineer", "engineering", "enter", "equity", "error", "escalation", "escape", "escape room", "estate", "ethereum", "etsy", "eurovision", "event", "excel", "exit", "expedia", "experience", "explainer", "exploit", "extrovert", "eyeliner", "face", "face id", "facebook", "facebook ads", "facepalm", "facetime", "fact", "fact checking", "fact-checking", "factoring", "factory", "fair", "fair play", "fake", "fake news", "fallback", "fanboy", "fancy", "fandom", "fanfic", "fangirl", "fanpage", "fantasy", "farming", "fashion", "fast", "fast fashion", "fast food", "feature", "feature flag", "feature request", "features", "featuring", "feed", "feedback", "fiction", "fig", "fight", "filter", "fine", "fine tuning", "fine-tuning", "finger", "finger food", "finisher", "fintech", "firebase", "firewall", "firmware", "fit", "fitbit", "fitness", "flag", "flash", "flash sale", "flex", "flexing", "flipchart", "flipping", "flow", "folk", "follow", "follow up", "follow-up", "follower", "followers", "fomo", "food", "food court", "food truck", "foodie", "football", "footer", "for", "for you page", "forbes", "forecast", "fork", "fortnite", "forward", "framework", "free", "free shipping", "free trial", "freelance", "freelancer", "freemium", "freeride", "freestyle", "freeware", "friday", "friend", "friend zone", "friends", "friends with benefits", "friendzone", "from", "front", "front office", "front-end", "frontend", "frontman", "frozen", "full", "full stack", "full time", "full-stack", "full-time", "fullstack", "fun", "funnel", "funny", "galaxy", "game", "game changer", "game of thrones", "game over", "game pass", "game show", "gamedev", "gamepad", "gameplay", "gamer", "gamification", "gaming", "gaming pc", "garage", "garmin", "gaslighting", "gate", "geforce", "gemini", "generation", "ghosting", "gift", "gift card", "girlboss", "girlfriend", "github", "github copilot", "gitlab", "giveaway", "glamping", "global", "global warming", "glow", "glow-up", "gmail", "go-live", "go-to", "go-to-market", "goat", "golang", "good", "good job", "good luck", "google", "google ads", "google analytics", "google cloud", "google docs", "google drive", "google maps", "google meet", "google pay", "google play", "google sheets", "gopro", "gps", "gpt", "grammy", "grammys", "graphic", "graphic designer", "graphql", "great", "great job", "green", "green flag", "greenwashing", "grind", "grinding", "grooming", "group", "group chat", "growth", "growth hacking", "growth mindset", "grunge", "guide", "gym", "hack", "hackathon", "hacker", "hacking", "halloween", "hand", "hands-on", "happy", "happy birthday", "happy end", "happy hour", "happy new year", "hard", "hard skills", "hardware", "harley", "harley-davidson", "harry", "harry potter", "hashtag", "hat", "hat trick", "hat-trick", "hate", "hate speech", "hater", "haul", "have", "headbanging", "header", "headhunter", "headless", "headline", "headliner", "headphones", "headquarters", "headset", "headshot", "healer", "healing", "health", "heart", "heatmap", "heavy", "heavy metal", "hello", "help", "help desk", "helpdesk", "hey", "hi-tech", "high", "high score", "high-tech", "highlighter", "highlights", "hiking", "hilton", "hip", "hip hop", "hip-hop", "hiring", "hiring manager", "hit", "hockey", "home", "home office", "homepod", "homestay", "honey", "hoodie", "hooks", "hop", "hosting", "hot", "hot dog", "hot-dog", "hotfix", "hotspot", "hour", "house", "housekeeping", "housewarming", "hover", "hubspot", "hugging", "hugging face", "hustle", "hut", "hybrid", "hybrid work", "hype", "hypebeast", "hyped", "icloud", "iconic", "imac", "inbound", "inbox", "incel", "inclusion", "inclusive", "incubator", "indeed", "indesign", "indie", "indiegogo", "indoor", "influencer", "influencer marketing", "info", "inheritance", "input", "insight", "insights", "instagram", "integration", "intelligence", "interface", "interior", "interior design", "internet", "internet of things", "internship", "interview", "introvert", "invite", "ipad", "iphone", "ipod", "issue", "jailbreak", "jam", "jam session", "james", "james bond", "java", "javascript", "jeans", "jeep", "jet", "jet lag", "jingle", "jir", "jira", "job", "job interview", "job offer", "jogging", "joker", "journaling", "joystick", "jungle", "k-pop", "key", "key account", "key account manager", "keyboard", "keynote", "kick-off", "kickboxing", "kickoff", "kickstarter", "kids", "kindle", "king", "kitesurfing", "knockout", "know-how", "kubernetes", "lag", "landing", "landing page", "language", "large", "large language model", "last", "last minute", "late", "late check-out", "latency", "launch", "layoff", "layoffs", "layout", "lead", "lead generation", "leader", "leaderboard", "leadership", "league", "learning", "leasing", "legacy", "legacy code", "lemonade", "lessons", "let's", "let's go", "level", "level up", "license", "life", "life hack", "lifehack", "lifestyle", "light", "light mode", "like", "line-up", "lineup", "link", "link in bio", "linkedin", "linter", "literally", "live", "live chat", "live coding", "live stream", "livestream", "lo-fi", "load", "load balancer", "lockdown", "lofi", "log", "log out", "logout", "long", "long read", "look", "loot", "lounge", "love", "love bombing", "low-code", "luck", "lunch", "lunchbox", "macbook", "macbook air", "macbook pro", "machine", "machine learning", "machinekind", "mail", "mailchimp", "mailing", "main", "mainstream", "make-up", "makeover", "maker", "makeup", "making", "making of", "malware", "management", "manager", "mansplaining", "maps", "market", "marketing", "marketing automation", "marketplace", "marvel", "master", "mastercard", "masterchef", "masterclass", "mastering", "match", "mcdonald's", "mcdonalds", "meal", "meal prep", "media", "meet", "meeting", "meeting room", "meetup", "member", "membership", "meme", "mental", "mental health", "mention", "merch", "merchandise", "merge", "merge request", "merry", "merry christmas", "message", "messenger", "metal", "metaverse", "mic", "microblading", "microservices", "microsoft", "microsoft office", "microsoft teams", "mid", "midjourney", "miles", "milestone", "milkshake", "millennials", "mindfulness", "mindset", "minecraft", "minions", "minute", "mixing", "mobile", "mobile app", "mock", "mockup", "mode", "model", "monday", "monetization", "monthly", "mood", "moodboard", "moshpit", "mousepad", "mouth", "muffin", "multiplayer", "multitasking", "music", "music video", "must", "must have", "must-have", "must-read", "must-see", "must-watch", "nearshoring", "nerf", "netflix", "network", "networking", "neural", "neural network", "new", "news", "newsletter", "newsroom", "next.js", "nice", "nice to have", "nickname", "nightclub", "nightlife", "nike", "no comment", "no problem", "no way", "no-code", "node.js", "nomad", "noob", "nordic", "nordic walking", "notebook", "notes", "notification", "notifications", "notion", "nuggets", "nvidia", "oasis", "off-road", "off-site", "offboarding", "offer", "office", "office manager", "offline", "offshoring", "offside", "ok boomer", "on-call", "on-site", "onboarding", "one-liner", "one-on-one", "onedrive", "online", "onlyfans", "open", "open mic", "open source", "open space", "open world", "open'er", "openai", "opt-in", "opt-out", "oscars", "out", "out of office", "outage", "outbound", "outcome", "outdoor", "outfit", "outlet", "outlook", "output", "outreach", "outsourcing", "over", "overbooking", "overclocking", "overengineering", "overflow", "overhead", "override", "oversize", "overthinking", "overtime", "overview", "overwatch", "owner", "ownership", "pace", "pack", "packaging", "page", "pain", "pain point", "paintball", "pair", "pair programming", "pancake", "pancakes", "part", "part time", "part-time", "partner", "partner in crime", "party", "pass", "password", "password manager", "past", "past simple", "paste", "patch", "path", "patreon", "patterns", "pay", "payload", "paypal", "paywall", "peeling", "pendrive", "pentest", "penthouse", "perfect", "performance", "performance marketing", "performance review", "perplexity", "personal", "personal brand", "personal branding", "personal trainer", "phishing", "photoshop", "pinterest", "pipeline", "pitch", "pitch deck", "pivot", "pixar", "pizza", "pizza hut", "placeholder", "plan", "plank", "planner", "planning", "plant-based", "play", "play store", "play-off", "playback", "playground", "playlist", "playoff", "playoffs", "playstation", "please", "plot", "plot twist", "plug-in", "plugin", "podcast", "podcaster", "point", "points", "policy", "pop-up", "popcorn", "popup", "positive", "post-it", "postmortem", "potter", "power", "power bank", "power bi", "power button", "powerbank", "powerpoint", "practice", "practices", "pre-order", "pre-workout", "premier", "premier league", "preorder", "prep", "prequel", "present", "present perfect", "pricing", "pride", "prime", "prime video", "print", "print screen", "printscreen", "privacy", "privacy policy", "pro", "problem", "prod", "producer", "product", "product designer", "product manager", "product owner", "production", "programming", "project", "project manager", "promo", "promo code", "prompt", "prompt engineer", "prompt engineering", "proof", "proof of concept", "protein", "proxy", "pub", "pull", "pull request", "push", "push notification", "python", "q&a", "qr", "quarterly", "queen", "quest", "queue", "quick", "quick win", "quiz", "radiohead", "rafting", "random", "ranked", "ransomware", "rate", "rave", "razer", "reach", "react", "reaction", "reactions", "read", "real", "real estate", "real-time", "reality", "reality show", "rebase", "rebranding", "recovery", "recruiter", "red", "red bull", "red carpet", "red flag", "reddit", "reebok", "reel", "reels", "refactor", "refactoring", "refinement", "refund", "release", "release notes", "remake", "remaster", "reminder", "remote", "remote work", "reply", "reply all", "repo", "report", "reporting", "repository", "repost", "request", "research", "resell", "reseller", "reset", "respawn", "response", "responsive", "restart", "retention", "retreat", "retrospective", "retweet", "revenue", "review", "reviewer", "ribs", "rihanna", "rizz", "road", "road trip", "roadie", "roadmap", "roadtrip", "roaming", "roast", "roblox", "rock'n'roll", "roguelike", "rollback", "rollercoaster", "rollout", "rolls-royce", "romance", "rooftop", "room", "router", "routine", "royale", "royalty", "rugby", "running", "runtime", "rush", "rust", "ryanair", "safe", "safe space", "sale", "sales", "sales manager", "salesforce", "samsung", "sandbox", "sandwich", "scalable", "scale-up", "scenes", "sci-fi", "science", "science fiction", "scientist", "scope", "scope creep", "score", "scraping", "screen", "screen time", "screenshot", "script", "scroll", "scrum", "scrum master", "search", "search engine", "second", "second hand", "second-hand", "see", "see you", "seen", "self-care", "self-help", "self-love", "self-service", "selfie", "selfie stick", "seller", "sequel", "server", "serverless", "service", "service desk", "session", "setup", "shake", "shaming", "shareholder", "sharepoint", "sharing", "shazam", "sheet", "sheets", "shein", "shell", "shipping", "shooter", "shop", "shopify", "shopping", "shortcut", "shorts", "shot", "shout-out", "shoutout", "show", "shower", "showroom", "showstopper", "side", "side hustle", "side project", "sign", "sign in", "sign up", "sign-off", "signal", "silicon", "silicon valley", "simp", "simple", "sitcom", "situationship", "six", "six pack", "sixpack", "skateboard", "skateboarding", "skatepark", "skater", "sketch", "skill", "skills", "skillset", "skin", "skincare", "skype", "skyscanner", "slack", "slasher", "slay", "slide", "slider", "slides", "small", "small talk", "smart", "smart city", "smart contract", "smart home", "smart tv", "smart watch", "smartband", "smartwatch", "smoothie", "snapchat", "sneakerhead", "sneakers", "snippet", "snooker", "snowboar", "snowboard", "snowboarding", "social", "social engineering", "social media manager", "soft", "soft launch", "soft skills", "software", "software developer", "software engineer", "software house", "software update", "songwriter", "sorry", "soul", "soulmate", "sound", "soundbar", "soundcheck", "soundtrack", "source", "space", "spacex", "spaghetti", "spaghetti code", "spam", "spawn", "speaker", "speech", "speed", "speed dating", "speedrun", "speedrunner", "spiderman", "spin-off", "spoiler", "spoiler alert", "spongebob", "sponsored", "spotify", "spreadsheet", "sprint", "sprint planning", "spyware", "squad", "squash", "stack", "stack overflow", "stack trace", "staging", "stakeholder", "stakeholders", "stalker", "stalking", "stand-up", "stand-up comedy", "standup", "star wars", "starbucks", "starlink", "start-up", "startup", "staycation", "steakhouse", "steam", "steampunk", "stick", "sticker", "stitch", "stopover", "storage", "store", "stories", "story", "story points", "storyboard", "storytelling", "streak", "stream", "streamer", "streaming", "street", "street food", "streetball", "streetstyle", "streetwear", "stretching", "study", "stunt", "stuntman", "stylist", "subscribe", "subscriber", "subscription", "subtitles", "subway", "subwoofer", "success", "summary", "summer", "summit", "sunrise", "sunset", "super", "super bowl", "supercharger", "superfood", "superhero", "superman", "support", "surfing", "surround", "survival", "sustainability", "sustainable", "swag", "swagger", "sweet", "sweetheart", "swift", "swipe", "swipe up", "sync", "t-shirt", "tagline", "take", "take away", "takeaway", "talent", "talent acquisition", "talent show", "talk", "talk show", "target", "task", "tea", "team", "team building", "team lead", "team leader", "teambuilding", "teams", "teamwork", "teaser", "tech", "tech debt", "tech lead", "tech stack", "technical", "technical debt", "template", "tensorflow", "test", "test case", "tests", "thank", "thank you", "thanks", "thanksgiving", "the", "theme", "things", "think", "think-tank", "this", "thread", "threads", "thriller", "thrones", "thumbnail", "thumbs", "thumbs up", "ticket", "tie-break", "tiktok", "tiktoker", "time", "time management", "time to market", "timeline", "timer", "tinder", "to do list", "to go", "to-do", "to-do list", "toggle", "token", "toner", "toolkit", "tooltip", "touch", "touch id", "touchpad", "touchscreen", "trace", "track", "tracking", "trackpad", "trader", "trading", "traffic", "trailer", "trainee", "trainer", "training", "trance", "transformer", "trekking", "trello", "trending", "trendsetter", "trial", "trick", "trip", "truck", "true", "true crime", "tuning", "tutorial", "tweet", "twerking", "twist", "twitch", "twitter", "typescript", "typing", "u2", "uber", "uber eats", "unboxing", "underdog", "underground", "unfollow", "unicorn", "unit", "unit test", "unit tests", "unity", "unplugged", "unsubscribe", "update", "upgrade", "upsell", "uptime", "use", "use case", "user", "user experience", "user interface", "user story", "username", "valley", "valorant", "vans", "venture", "venture capital", "vibe", "vibes", "video", "video call", "views", "vintage", "viral", "virtual", "virtual reality", "vision", "vlog", "vlogger", "voice", "voice message", "voucher", "walking", "wallet", "warming", "waste", "watch", "watching", "waterfall", "way", "waze", "wearable", "wearables", "web", "web app", "web designer", "web3", "webcam", "webhook", "wednesday", "weekday", "weeken", "weekend", "weekendowych", "weekly", "weird", "welcome", "well", "well done", "well-being", "wellbeing", "wellness", "whatever", "whatsapp", "whey", "whiteboard", "whitepaper", "wi-fi", "widget", "wifi", "wimbledon", "win-win", "windows", "windsurfing", "wireframe", "wireless", "wishlist", "with", "woke", "woodstock", "word", "word of mouth", "wordpress", "work from home", "work-life", "work-life balance", "workaround", "workout", "workshop", "world", "wow", "wrap", "xbox", "year", "yes", "yolo", "you", "youtube", "youtube shorts", "youtuber", "zero", "zero waste", "zone"], "read_as_polish": ["art", "bank", "blog", "bloga", "budżet", "canva", "data", "demo", "duolingo", "figma", "figmie", "film", "golf", "grill", "hotel", "internet", "jazz", "kebab", "komputer", "laptop", "laptopa", "laptopem", "laptopie", "lider", "list", "lobby", "lunchbox", "mecz", "media", "menu", "metal", "model", "ok", "partner", "pizza", "plan", "premier", "pro", "problem", "radio", "retro", "rock", "serial", "smartfon", "sms", "sport", "star", "stop", "super", "sushi", "tablet", "talent", "taxi", "telegram", "tenis", "test", "tour", "trener", "video", "webinar", "webinaru", "webinarze", "zero", "zoom"], "suffixes": ["ami", "ach", "owi", "iem", "owy", "owa", "owe", "em", "om", "ów", "ie", "ze", "u", "a", "y", "i", "e", "ę"], "loan_stems": ["mail", "weekend", "facebook", "lunch", "router", "link", "youtube", "google", "instagram", "podcast", "playlist", "meeting", "smartphone", "deadline", "feedback", "update", "spotify", "netflix", "whatsapp", "messenger", "tiktok", "linkedin", "iphone"], "pl_inflected_block": ["add-ona", "add-one", "add-oni", "airbusa", "all-stara", "all-stare", "all-stary", "ally", "andy", "andów", "angela", "angeli", "apple musicie", "arte", "bankach", "bankami", "banki", "bankiem", "bankowe", "bankowy", "banku", "banków", "batmana", "bestie", "billboardy", "bitcoina", "bonda", "bossa", "brandy", "breaking newsa", "breaking newsami", "breaking newsy", "breaking newsów", "brexitu", "bugu", "burger kinga", "burgera", "burgery", "business angela", "business angeli", "business casey", "business modela", "business modelach", "business modele", "business modelem", "business modeli", "business modelu", "business partnera", "business partnerami", "business partnerem", "business partnerów", "business planach", "business planami", "business planem", "business planie", "business planom", "business planu", "business plany", "business planów", "casey", "catch-upa", "chata", "chaty", "chatę", "chilli", "cold maila", "cold maile", "cold mailem", "cold maili", "cold mailowy", "cold mailu", "content marketingowe", "content marketingowy", "content marketingu", "covery", "cross-platforma", "cross-platformach", "cross-platformie", "cross-platformy", "cross-platformę", "deep worka", "deep worki", "deep workiem", "deep worku", "deep worków", "della", "disneya", "doga", "dogi", "dragi", "dream teamu", "e-maila", "e-maile", "e-mailem", "e-maili", "e-mailowy", "e-mailu", "edge casey", "eventy", "face ida", "face ide", "face idę", "fair playa", "fairy", "fake newsa", "fake newsami", "fake newsy", "fake newsów", "flaga", "flagami", "flagi", "flagę", "follow upa", "follow-upa", "fora", "forach", "forów", "frontach", "frontem", "frontowe", "frontu", "fronty", "frontów", "game passe", "glowa", "glowe", "glowie", "glowy", "go-to-marketach", "go-to-marketu", "go-toi", "go-toy", "good lucky", "google playa", "green flaga", "green flagami", "green flagi", "green flagę", "greene", "group chata", "group chaty", "group chatę", "hands-ona", "hands-one", "hands-oni", "hardy", "harry pottera", "hashtagi", "hashtagiem", "hashtagu", "hate", "heavy metale", "heavy metali", "heavy metalowa", "heavy metalowe", "heavy metalowy", "heavy metalu", "help deska", "help deskach", "help deski", "help deskę", "hip hope", "hip hopu", "hip-hope", "hip-hopu", "hitem", "hitu", "hity", "hitów", "hope", "hopu", "huta", "huty", "internetach", "internetem", "internetowa", "internetowe", "internetowi", "internetowy", "internetu", "internety", "internetów", "ipada", "james bonda", "jamesa", "jamesem", "jamie", "jamy", "jeansy", "jokera", "k-pope", "k-popu", "kinga", "kingi", "know-howie", "large language modele", "large language modeli", "large language modelu", "let's goi", "level upa", "line-upa", "live chata", "live chaty", "live chatę", "loga", "logiem", "long ready", "lucky", "maine", "mainstreamowe", "mainstreamu", "make-upa", "making ofe", "managera", "managerem", "marketach", "marketingiem", "marketingowa", "marketingowe", "marketingowy", "marketingu", "marketu", "marvela", "metale", "metali", "metalowa", "metalowe", "metalowy", "metalu", "microsoftu", "midi", "modela", "modelach", "modele", "modelem", "modeli", "modelu", "musicie", "must-ready", "newsa", "newsami", "newsy", "newsów", "nicei", "no problemach", "no problemami", "no problemem", "no problemie", "no problemom", "no problemu", "no problemy", "no problemów", "parti", "partie", "partnera", "partnerami", "partnerem", "partnerów", "party", "passe", "passę", "pasta", "pasty", "pastę", "performance marketingowe", "performance marketingowy", "performance marketingu", "photoshopie", "pizza huta", "pizza huty", "planach", "planami", "planem", "planie", "planom", "planu", "plany", "planów", "playa", "pop-upa", "popcornu", "pottera", "power bankach", "power bankami", "power banki", "power bankiem", "power bankowe", "power bankowy", "power banku", "power banków", "power bie", "pre-orderu", "premiera", "premierem", "premierowi", "premiery", "premierze", "premierów", "premierę", "problemach", "problemami", "problemem", "problemie", "problemom", "problemu", "problemy", "problemów", "pubie", "pubu", "q&aa", "q&ae", "q&ai", "q&au", "quick wina", "quick winie", "quick winy", "quick winę", "ready", "realem", "realu", "red flaga", "red flagami", "red flagi", "red flagę", "reda", "samsunga", "screena", "screeny", "service deska", "service deskach", "service deski", "service deskę", "spamu", "stand-upa", "start-upa", "supermana", "talentem", "talentu", "talenty", "talentów", "teamu", "test casey", "testach", "testami", "testem", "testowy", "testu", "testy", "testów", "they", "time to marketach", "time to marketu", "to do lista", "to do listach", "to do listami", "to do liste", "to do listem", "to do listu", "to do listy", "to do listów", "to do listę", "to goi", "to-do lista", "to-do listach", "to-do listami", "to-do liste", "to-do listem", "to-do listu", "to-do listy", "to-do listów", "to-do listę", "to-doi", "to-doom", "touch ida", "touch ide", "touch idę", "tweeta", "tweetach", "tweetem", "tweety", "tweetów", "twittera", "twitterem", "twitterze", "unit testach", "unit testem", "unit testu", "unit testy", "unit testów", "unity", "use casey", "win-wina", "win-wine", "win-winem", "win-wini", "win-winie", "win-winy", "win-winę", "windowsa"]}

english/english_detect.py

@@ -0,0 +1,129 @@
+#!/usr/bin/env python3
+"""
+Find English words in ordinary Polish text and wrap them in braces for polish_mixed_cleaners (run time, ~µs per word).
+
+A word (or a Polish-inflected form of it: meetingu -> meeting + u) is English when
+  - it is in the curated list (english_words.txt: the English words of the training set, brands, common loanwords), or
+  - it is common in English and clearly rarer in Polish (wordfreq Zipf: en >= 3, en - pl >= MARGIN) and not a common
+    Polish word (pl < 4.5: protects "to", "dom", "pies"…), and it has no Polish diacritics;
+and it is not on the read-as-Polish list (laptop, retro, film…). Explicit braces in the input are kept as they are.
+
+  from english_detect import mark_english;  mark_english("Po meetingu wyślę ci feedback.")  ->  "Po {meeting}u wyślę ci {feedback}."
+  python3 voxcpm/english_detect.py --eval data_v3/matcha_fav2/metadata.csv
+"""
+import re
+from functools import lru_cache
+from pathlib import Path
+
+HERE = Path(__file__).resolve().parent
+MARGIN, EN_MIN, PL_MAX = 1.5, 3.5, 3.5
+PL_INFLECTED_MAX = 3.0   # an inflected form at least this common in Polish is a Polish word
+SUFFIXES = sorted(["ami", "ach", "owi", "iem", "em", "om", "ów", "ie", "u", "a", "y", "i", "e", "ę", "ze", "owy", "owa", "owe"], key=len, reverse=True)
+READ_AS_POLISH = {"retro", "webinar", "duolingo", "lobby", "laptop", "blog", "budżet", "zoom", "canva", "figma", "film", "plan", "super", "stop",
+                  "sport", "hotel", "tenis", "golf", "mecz", "lider", "trener", "komputer", "internet", "radio", "taxi", "menu", "pizza",
+                  "smartfon", "tablet", "serial", "lunchbox", "jazz", "rock", "sms", "ok", "demo", "sushi", "kebab", "grill", "telegram",
+                  "webinaru", "webinarze", "laptopa", "laptopie", "laptopem", "bloga", "figmie"} | {"art", "bank", "data", "internet", "list", "media", "metal", "model", "partner", "pizza", "plan", "premier", "pro", "problem", "star", "super", "talent", "test", "tour", "video", "zero"}
+PL_HOMOGRAPHS = {"to", "a", "i", "na", "do", "on", "no", "nie", "ale", "jak", "tak", "most", "drug", "wind", "now", "got", "but", "star", "form",
+                 "set", "lot", "men", "end", "car", "list", "post", "tor", "pan", "pal", "bar", "kit", "mama", "data", "kino", "wars", "win",
+                 "desk", "legend", "work"}
+# established loans whose inflected forms are frequent in Polish text (maila, weekendu, Facebooku) but still English-sounding
+LOAN_STEMS = ("mail", "weekend", "facebook", "lunch", "router", "link", "youtube", "google", "instagram", "podcast", "playlist", "meeting",
+              "smartphone", "deadline", "feedback", "update", "spotify", "netflix", "whatsapp", "messenger", "tiktok", "linkedin", "iphone")
+PL_DIACRITICS = re.compile(r"[ąćęłńóśźż]", re.I)
+WORD = re.compile(r"\{[^{}]*\}|[A-Za-zÀ-ÿĄĆĘŁŃÓŚŹŻąćęłńóśźż][A-Za-zÀ-ÿĄĆĘŁŃÓŚŹŻąćęłńóśźż'’-]*")
+
+
+def _load_curated():
+    f = HERE / "english_words.txt"
+    if not f.exists():
+        return set()
+    out = set()
+    for l in f.read_text(encoding="utf-8").splitlines():
+        l = l.strip().lower()
+        if l and not l.startswith("#"):
+            out.add(l)
+            out |= {w for w in l.split() if len(w) > 2 and w not in PL_HOMOGRAPHS}   # words of phrases: "pull request" -> pull, request
+    return out
+
+
+CURATED = _load_curated()
+
+
+def known(w):
+    lw = w.lower()
+    return lw in CURATED and lw not in READ_AS_POLISH
+
+
+@lru_cache(maxsize=200000)
+def by_frequency(w):
+    """fallback for whole words not on the list: clearly English, rare in Polish"""
+    lw = w.lower()
+    if lw in READ_AS_POLISH or PL_DIACRITICS.search(lw) or len(lw) < 4:
+        return False
+    try:
+        from wordfreq import zipf_frequency
+    except ImportError:
+        return False
+    en, pl = zipf_frequency(lw, "en"), zipf_frequency(lw, "pl")
+    return en >= EN_MIN and en - pl >= MARGIN and pl < PL_MAX
+
+
+@lru_cache(maxsize=200000)
+def pl_word(lw):
+    try:
+        from wordfreq import zipf_frequency
+    except ImportError:
+        return False
+    return zipf_frequency(lw, "pl") >= PL_INFLECTED_MAX
+
+
+def is_english(w):
+    return known(w) or by_frequency(w)
+
+
+def split_word(w):
+    """English stem + Polish ending, or None"""
+    if is_english(w):
+        return w, ""
+    lw = w.lower()
+    if "'" in w or "’" in w:                                       # deadline'u, pendrive'a
+        st = w.replace("’", "'").split("'", 1)
+        if known(st[0]) or by_frequency(st[0]):
+            return st[0], "'" + st[1] if st[1] else ""
+    if pl_word(lw) and not any(lw.startswith(x) and len(lw) - len(x) <= 3 for x in LOAN_STEMS):
+        return None      # wina, legendy, deskę, worki: Polish words whose stem happens to be English
+    for k in SUFFIXES:   # inflected forms only of listed words (stripping endings of arbitrary words: nowy -> "now" + y)
+        if lw.endswith(k) and len(w) - len(k) >= 3 and known(w[:-len(k)]):
+            return w[:-len(k)], w[-len(k):]
+    return None
+
+
+def mark_english(text):
+    def rep(m):
+        w = m.group(0)
+        if w.startswith("{"):
+            return w
+        parts = [p for p in re.split(r"(-)", w)] if "-" in w and not is_english(w) else [w]   # Wi-Fi stays whole if known
+        if len(parts) > 1:
+            return w if split_word(w) is None else "{" + split_word(w)[0] + "}" + split_word(w)[1]
+        s = split_word(w)
+        return w if s is None else "{" + s[0] + "}" + s[1]
+    out = WORD.sub(rep, text)
+    return re.sub(r"\}(\s+)\{", r"\1", out)      # adjacent English words -> one span: {pull request}
+
+
+if __name__ == "__main__":
+    import argparse, csv
+    ap = argparse.ArgumentParser(); ap.add_argument("--eval", type=Path); ap.add_argument("--show", type=int, default=15); a = ap.parse_args()
+    rows = list(csv.DictReader(a.eval.open(encoding="utf-8"), delimiter="|"))
+    tp = fp = fn = 0; fps, fns = [], []
+    for r in rows:
+        gold = {s.lower() for s in re.findall(r"\{([^}]*)\}", r["marked"])}
+        pred = {s.lower() for s in re.findall(r"\{([^}]*)\}", mark_english(r["text"]))}
+        gw = {w for s in gold for w in s.split()}; pw = {w for s in pred for w in s.split()}
+        tp += len(gw & pw); fp += len(pw - gw); fn += len(gw - pw)
+        fps += [(w, r["text"]) for w in pw - gw]; fns += [(w, r["text"]) for w in gw - pw]
+    print(f"words: precision {tp / max(1, tp + fp):.3f}, recall {tp / max(1, tp + fn):.3f} (tp {tp}, fp {fp}, fn {fn})")
+    from collections import Counter
+    print("most common false positives:", Counter(w for w, _ in fps).most_common(a.show))
+    print("most common misses:", Counter(w for w, _ in fns).most_common(a.show))

english/english_words.txt

@@ -0,0 +1,1549 @@
+# English words and phrases pronounced the English way inside Polish speech
+# + English words of the fav2 training set (2026-10-02); Polish-pronounced ones are in english_detect.READ_AS_POLISH
+ac/dc
+accelerator
+account manager
+accountability
+achievement
+action points
+add-on
+adobe
+aesthetic
+affiliate
+after party
+afterparty
+afterwork
+agile
+airbnb
+airbus
+airdrop
+airplay
+airpods
+airtag
+algorithm
+aliexpress
+alignment
+all inclusive
+all-star
+alternative
+amazing
+amazon
+amazon prime
+analytics
+angular
+anyway
+app
+app store
+apple
+apple music
+apple pay
+apple store
+apple tv
+apple watch
+apps
+art director
+asap
+attachment
+audiobook
+augmented reality
+autoplay
+awesome
+azure
+babe
+baby boomers
+baby shower
+babysitter
+back office
+back-end
+backdoor
+backend
+background
+backlog
+backpack
+backpacker
+backpacking
+backstage
+backup
+bacon
+badge
+bae
+barbecue
+barber
+barbershop
+barbie
+bartender
+baseball
+basic
+basically
+basketball
+batman
+battle royale
+beatbox
+beauty
+beef
+behind the scenes
+benchmark
+benchmarking
+best friend
+best practice
+best practices
+best seller
+bestie
+big brother
+big data
+billboard
+binge watching
+binge-watching
+bitcoin
+black friday
+blame
+blind date
+bling
+blockbuster
+blockchain
+blocker
+blogger
+bluetooth
+boarding
+body positive
+body shaming
+bodybuilder
+bodybuilding
+boeing
+boilerplate
+bolt
+booking
+boomer
+booster
+bootcamp
+boss fight
+bottleneck
+bowl
+bowling
+box office
+boyfriend
+boysband
+brainstorm
+brainstorming
+branch
+brand
+brand ambassador
+brand awareness
+brandbook
+branding
+breakdance
+breakdown
+breakfast
+breaking
+breaking news
+breakpoint
+brexit
+brief
+briefing
+brownie
+brunch
+bubble
+bubble tea
+bucket list
+budget
+bug
+bug bounty
+bug report
+bugfix
+bugs
+build
+bumble
+bundle
+burger
+burger king
+burnout
+burpees
+business
+business angel
+business case
+business intelligence
+business model
+business partner
+business plan
+button
+buzz
+buzzword
+by the way
+bye
+c-level
+cache
+calisthenics
+call
+call to action
+cancel culture
+capcut
+cappuccino
+caption
+car sharing
+cardio
+career path
+carry
+carsharing
+case study
+cash
+cash flow
+cashback
+casual
+casual friday
+catcalling
+catch-up
+catering
+challenge
+champions league
+changelog
+chargeback
+charger
+chat
+chat gpt
+chatbot
+chatgpt
+cheat
+cheat sheet
+check-in
+check-out
+checkbox
+checklist
+checkout
+checkpoint
+cheers
+cheeseburger
+cheesecake
+chill
+chillout
+christmas
+chromebook
+churn
+churn rate
+cider
+city break
+claim
+clean code
+clearance
+clickbait
+clickup
+cliffhanger
+climate change
+clone
+close friends
+closing
+cloud
+cloud computing
+cloudflare
+cluster
+clutch
+co-host
+co-op
+coach
+coachella
+coaching
+code
+code review
+codebase
+coder
+coding
+coffee shop
+cold call
+cold mail
+coldplay
+coliving
+collab
+collabs
+comeback
+comedy
+coming out
+coming-of-age
+commit
+commitment
+community manager
+compiler
+compliance
+component
+computer vision
+concealer
+conference call
+confluence
+congrats
+congratulations
+consent
+contactless
+container
+content
+content creator
+content marketing
+continue
+continuous integration
+converse
+conversion rate
+cookie
+cookies
+cool
+copilot
+copy
+copy paste
+copy-paste
+copyright
+copywriter
+copywriting
+corporate
+cosmopolitan
+cosplay
+counter-strike
+country
+coupon
+cover
+coworking
+cozy
+craft
+creative director
+creator
+credit card
+creepy
+crew
+cringe
+cronjob
+cross-platform
+cross-sell
+crossfit
+crossover
+crowdfunding
+crunch
+crush
+crypto
+cupcake
+curling
+customer experience
+customer service
+customer success
+customer support
+cut-off
+cute
+cutscene
+cyber monday
+cyberbullying
+cyberpunk
+daily
+daily standup
+dark mode
+darling
+darts
+dashboard
+data analyst
+data engineer
+data science
+data scientist
+dataset
+date
+dating
+deadlift
+deadline
+deadlock
+deadpool
+deal
+dealer
+debriefing
+debug
+debugger
+debugging
+deck
+deep dive
+deep learning
+deep work
+deepfake
+deepseek
+deliverable
+delivery
+dell
+dependency
+deploy
+deployment
+design
+design patterns
+designer
+desktop
+dev
+developer
+devops
+devtools
+diff
+digital detox
+digital nomad
+dinner
+dip
+direct message
+director
+discord
+discount
+discovery
+disney
+disneyland
+diss
+diversity
+dj
+docker
+donut
+double click
+downgrade
+downtime
+draft
+drag and drop
+dream job
+dream team
+dress code
+drip
+drive
+drive-in
+drive-thru
+drop
+dropbox
+dropdown
+dropshipping
+dubstep
+duty-free
+déjà vu
+e-book
+e-commerce
+e-learning
+e-mail
+earbuds
+early access
+early adopter
+early bird
+easter egg
+easy
+easyjet
+ebay
+ebook
+eco-friendly
+ecommerce
+edge case
+elasticsearch
+elearning
+elevator pitch
+email
+embedding
+eminem
+emirates
+emmy
+emoji
+emote
+employer branding
+empowerment
+encryption
+end-to-end
+endpoint
+engagement
+enter
+equity
+error
+escalation
+escape room
+ethereum
+etsy
+eurovision
+event
+excel
+exit
+expedia
+explainer
+exploit
+extrovert
+eyeliner
+face id
+facebook
+facebook ads
+facepalm
+facetime
+fact checking
+fact-checking
+factoring
+factory
+fair play
+fake
+fake news
+fallback
+fanboy
+fancy
+fandom
+fanfic
+fangirl
+fanpage
+fantasy
+farming
+fast fashion
+fast food
+feature
+feature flag
+feature request
+features
+featuring
+feed
+feedback
+fig
+filter
+fine tuning
+fine-tuning
+finger food
+finisher
+fintech
+firebase
+firewall
+firmware
+fit
+fitbit
+fitness
+flash sale
+flex
+flexing
+flipchart
+flipping
+flow
+folk
+follow
+follow up
+follow-up
+follower
+followers
+fomo
+food court
+food truck
+foodie
+football
+footer
+for you page
+forbes
+forecast
+fork
+fortnite
+forward
+framework
+free shipping
+free trial
+freelance
+freelancer
+freemium
+freeride
+freestyle
+freeware
+friday
+friend zone
+friends
+friends with benefits
+friendzone
+front office
+front-end
+frontend
+frontman
+frozen
+full stack
+full time
+full-stack
+full-time
+fullstack
+fun
+funnel
+funny
+galaxy
+game
+game changer
+game of thrones
+game over
+game pass
+game show
+gamedev
+gamepad
+gameplay
+gamer
+gamification
+gaming
+gaming pc
+garage
+garmin
+gaslighting
+gate
+geforce
+gemini
+ghosting
+gift card
+girlboss
+girlfriend
+github
+github copilot
+gitlab
+giveaway
+glamping
+global warming
+glow
+glow-up
+gmail
+go-live
+go-to
+go-to-market
+goat
+golang
+good job
+good luck
+google
+google ads
+google analytics
+google cloud
+google docs
+google drive
+google maps
+google meet
+google pay
+google play
+google sheets
+gopro
+gps
+grammy
+grammys
+graphic designer
+graphql
+great job
+green flag
+greenwashing
+grind
+grinding
+grooming
+group chat
+growth
+growth hacking
+growth mindset
+grunge
+guide
+gym
+hackathon
+hacker
+hacking
+halloween
+hands-on
+happy birthday
+happy end
+happy hour
+happy new year
+hard skills
+hardware
+harley
+harley-davidson
+harry potter
+hashtag
+hat trick
+hat-trick
+hate
+hate speech
+hater
+haul
+headbanging
+header
+headhunter
+headless
+headline
+headliner
+headphones
+headquarters
+headset
+headshot
+healer
+healing
+heart
+heatmap
+heavy metal
+hello
+help desk
+helpdesk
+hey
+hi-tech
+high score
+high-tech
+highlighter
+highlights
+hiking
+hilton
+hip hop
+hip-hop
+hiring manager
+hit
+hockey
+home office
+homepod
+homestay
+honey
+hoodie
+hooks
+hosting
+hot dog
+hot-dog
+hotfix
+hotspot
+house
+housekeeping
+housewarming
+hover
+hubspot
+hugging face
+hustle
+hybrid work
+hype
+hypebeast
+hyped
+icloud
+iconic
+imac
+inbound
+inbox
+incel
+inclusion
+incubator
+indeed
+indesign
+indie
+indiegogo
+indoor
+influencer
+influencer marketing
+info
+inheritance
+input
+insight
+insights
+instagram
+interface
+interior design
+internet of things
+internship
+introvert
+invite
+ipad
+iphone
+ipod
+issue
+jailbreak
+jam session
+james bond
+java
+javascript
+jeans
+jeep
+jet lag
+jingle
+jir
+jira
+job
+job interview
+job offer
+jogging
+joker
+journaling
+joystick
+jungle
+k-pop
+key account
+key account manager
+keyboard
+keynote
+kick-off
+kickboxing
+kickoff
+kickstarter
+kids
+kindle
+kitesurfing
+knockout
+know-how
+kubernetes
+landing
+landing page
+large language model
+last minute
+late check-out
+latency
+launch
+layoff
+layoffs
+layout
+lead
+lead generation
+leaderboard
+leadership
+leasing
+legacy
+legacy code
+lemonade
+lessons
+let's go
+level
+level up
+license
+life hack
+lifehack
+lifestyle
+light mode
+like
+line-up
+lineup
+link
+link in bio
+linkedin
+linter
+literally
+live
+live chat
+live coding
+live stream
+livestream
+lo-fi
+load balancer
+lockdown
+lofi
+log out
+logout
+long read
+look
+loot
+lounge
+love bombing
+low-code
+lunch
+lunchbox
+macbook
+macbook air
+macbook pro
+machine learning
+machinekind
+mail
+mailchimp
+mailing
+main
+mainstream
+make-up
+makeover
+maker
+makeup
+making of
+malware
+management
+manager
+mansplaining
+marketing automation
+marketplace
+marvel
+mastercard
+masterchef
+masterclass
+mastering
+match
+mcdonald's
+mcdonalds
+meal prep
+meeting
+meeting room
+meetup
+member
+membership
+meme
+mental health
+mention
+merch
+merchandise
+merge
+merge request
+merry christmas
+messenger
+metaverse
+microblading
+microservices
+microsoft
+microsoft office
+microsoft teams
+mid
+midjourney
+miles
+milestone
+milkshake
+millennials
+mindfulness
+mindset
+minecraft
+minions
+mixing
+mobile app
+mock
+mockup
+monetization
+monthly
+mood
+moodboard
+moshpit
+mousepad
+muffin
+multiplayer
+multitasking
+music video
+must have
+must-have
+must-read
+must-see
+must-watch
+nearshoring
+nerf
+netflix
+networking
+neural network
+news
+newsletter
+newsroom
+next.js
+nice
+nice to have
+nickname
+nightclub
+nightlife
+nike
+no comment
+no problem
+no way
+no-code
+node.js
+noob
+nordic walking
+notebook
+notification
+notifications
+notion
+nuggets
+nvidia
+oasis
+off-road
+off-site
+offboarding
+office
+office manager
+offline
+offshoring
+offside
+ok boomer
+on-call
+on-site
+onboarding
+one-liner
+one-on-one
+onedrive
+online
+onlyfans
+open mic
+open source
+open space
+open world
+open'er
+openai
+opt-in
+opt-out
+oscars
+out of office
+outage
+outbound
+outcome
+outdoor
+outfit
+outlet
+outlook
+output
+outreach
+outsourcing
+overbooking
+overclocking
+overengineering
+overhead
+override
+oversize
+overthinking
+overtime
+overview
+overwatch
+ownership
+pace
+packaging
+pain point
+paintball
+pair programming
+pancake
+pancakes
+part time
+part-time
+partner in crime
+party
+password
+password manager
+past simple
+patch
+patreon
+payload
+paypal
+paywall
+peeling
+pendrive
+pentest
+penthouse
+performance
+performance marketing
+performance review
+perplexity
+personal brand
+personal branding
+personal trainer
+phishing
+photoshop
+pinterest
+pipeline
+pitch
+pitch deck
+pivot
+pixar
+pizza hut
+placeholder
+plank
+planner
+planning
+plant-based
+play store
+play-off
+playback
+playground
+playlist
+playoff
+playoffs
+playstation
+please
+plot twist
+plug-in
+plugin
+podcast
+podcaster
+pop-up
+popcorn
+popup
+post-it
+postmortem
+power bank
+power bi
+power button
+powerbank
+powerpoint
+pre-order
+pre-workout
+premier league
+preorder
+prequel
+present perfect
+pricing
+pride
+prime video
+print screen
+printscreen
+privacy policy
+prod
+producer
+product designer
+product manager
+product owner
+production
+project manager
+promo code
+prompt
+prompt engineer
+prompt engineering
+proof of concept
+protein
+proxy
+pub
+pull
+pull request
+push
+push notification
+python
+q&a
+qr
+quarterly
+queen
+quest
+queue
+quick win
+quiz
+radiohead
+rafting
+random
+ranked
+ransomware
+rave
+razer
+reach
+react
+reaction
+reactions
+real estate
+real-time
+reality show
+rebase
+rebranding
+recovery
+recruiter
+red bull
+red carpet
+red flag
+reddit
+reebok
+reel
+reels
+refactor
+refactoring
+refinement
+refund
+release
+release notes
+remake
+remaster
+reminder
+remote
+remote work
+reply
+reply all
+repo
+reporting
+repository
+repost
+request
+research
+resell
+reseller
+reset
+respawn
+response
+responsive
+restart
+retention
+retreat
+retrospective
+retweet
+revenue
+review
+reviewer
+ribs
+rihanna
+rizz
+road trip
+roadie
+roadmap
+roadtrip
+roaming
+roast
+roblox
+rock'n'roll
+roguelike
+rollback
+rollercoaster
+rollout
+rolls-royce
+romance
+rooftop
+router
+routine
+royalty
+rugby
+running
+runtime
+rush
+rust
+ryanair
+safe space
+sales manager
+salesforce
+samsung
+sandbox
+sandwich
+scalable
+scale-up
+sci-fi
+science fiction
+scope creep
+scraping
+screen
+screen time
+screenshot
+script
+scroll
+scrum
+scrum master
+search engine
+second hand
+second-hand
+see you
+seen
+self-care
+self-help
+self-love
+self-service
+selfie
+selfie stick
+sequel
+server
+serverless
+service desk
+session
+setup
+shake
+shareholder
+sharepoint
+shazam
+sheet
+shein
+shell
+shooter
+shopify
+shopping
+shortcut
+shorts
+shot
+shout-out
+shoutout
+show
+showroom
+showstopper
+side hustle
+side project
+sign in
+sign up
+sign-off
+signal
+silicon valley
+simp
+sitcom
+situationship
+six pack
+sixpack
+skateboard
+skateboarding
+skatepark
+skater
+sketch
+skill
+skills
+skillset
+skin
+skincare
+skype
+skyscanner
+slack
+slasher
+slay
+slide
+slider
+slides
+small talk
+smart city
+smart contract
+smart home
+smart tv
+smart watch
+smartband
+smartwatch
+smoothie
+snapchat
+sneakerhead
+sneakers
+snippet
+snooker
+snowboar
+snowboard
+snowboarding
+social engineering
+social media manager
+soft launch
+soft skills
+software
+software developer
+software engineer
+software house
+software update
+songwriter
+sorry
+soul
+soulmate
+sound
+soundbar
+soundcheck
+soundtrack
+spacex
+spaghetti code
+spam
+spawn
+speaker
+speech
+speed dating
+speedrun
+speedrunner
+spiderman
+spin-off
+spoiler
+spoiler alert
+spongebob
+sponsored
+spotify
+spreadsheet
+sprint
+sprint planning
+spyware
+squad
+squash
+stack
+stack overflow
+stack trace
+staging
+stakeholder
+stakeholders
+stalker
+stalking
+stand-up
+stand-up comedy
+standup
+star wars
+starbucks
+starlink
+start-up
+startup
+staycation
+steakhouse
+steam
+steampunk
+sticker
+stitch
+stopover
+storage
+stories
+story
+story points
+storyboard
+storytelling
+streak
+stream
+streamer
+streaming
+street food
+streetball
+streetstyle
+streetwear
+stretching
+stunt
+stuntman
+stylist
+subscribe
+subscriber
+subscription
+subtitles
+subway
+subwoofer
+summary
+summer
+summit
+sunrise
+sunset
+super bowl
+supercharger
+superfood
+superhero
+superman
+support
+surfing
+surround
+survival
+sustainability
+sustainable
+swag
+swagger
+sweet
+sweetheart
+swift
+swipe
+swipe up
+sync
+t-shirt
+tagline
+take away
+takeaway
+talent acquisition
+talent show
+talk
+talk show
+target
+task
+team
+team building
+team lead
+team leader
+teambuilding
+teams
+teamwork
+teaser
+tech
+tech debt
+tech lead
+tech stack
+technical debt
+template
+tensorflow
+test case
+thank you
+thanks
+thanksgiving
+theme
+think
+think-tank
+this
+thread
+threads
+thriller
+thumbnail
+thumbs up
+ticket
+tie-break
+tiktok
+tiktoker
+time management
+time to market
+timeline
+timer
+tinder
+to do list
+to go
+to-do
+to-do list
+toggle
+token
+toner
+toolkit
+tooltip
+touch id
+touchpad
+touchscreen
+track
+tracking
+trackpad
+trader
+trading
+traffic
+trailer
+trainee
+training
+trance
+transformer
+trekking
+trello
+trending
+trendsetter
+trial
+true crime
+tutorial
+tweet
+twerking
+twitch
+twitter
+typescript
+typing
+u2
+uber
+uber eats
+unboxing
+underdog
+underground
+unfollow
+unicorn
+unit test
+unit tests
+unity
+unplugged
+unsubscribe
+update
+upgrade
+upsell
+uptime
+use case
+user
+user experience
+user interface
+user story
+username
+valorant
+vans
+venture capital
+vibe
+vibes
+video call
+views
+vintage
+viral
+virtual reality
+vlog
+vlogger
+voice message
+voucher
+wallet
+waterfall
+waze
+wearable
+wearables
+web app
+web designer
+web3
+webcam
+webhook
+wednesday
+weekday
+weeken
+weekend
+weekendowych
+weekly
+weird
+welcome
+well done
+well-being
+wellbeing
+wellness
+whatever
+whatsapp
+whey
+whiteboard
+whitepaper
+wi-fi
+widget
+wifi
+wimbledon
+win-win
+windows
+windsurfing
+wireframe
+wireless
+wishlist
+woke
+woodstock
+word
+word of mouth
+wordpress
+work from home
+work-life
+work-life balance
+workaround
+workout
+workshop
+wow
+wrap
+xbox
+yes
+yolo
+youtube
+youtube shorts
+youtuber
+zero waste

english/polish_cleaners.py

@@ -0,0 +1,85 @@
+"""
+Polski cleaner dla Matcha-TTS. scripts/patch_matcha.py dokleja go do matcha/text/cleaners.py.
+
+Wymaga: libespeak-ng (apt: libespeak-ng1 + espeak-ng-data; binarka espeak-ng NIE jest potrzebna),
+phonemizer (jest w deps Matcha). Jeśli phonemizer nie znajdzie biblioteki, ustaw
+PHONEMIZER_ESPEAK_LIBRARY=/usr/lib/aarch64-linux-gnu/libespeak-ng.so.1
+
+Wyjście: IPA z akcentami – ten sam format co english_cleaners2. Polskie samogłoski nosowe
+(ɔ̃, ɛ̃) używają COMBINING TILDE U+0303, którego nie ma w oryginalnym symbols.py –
+patch_matcha.py dokłada go do _letters_ipa (n_vocab 178 -> 179).
+"""
+import re
+from phonemizer.backend import EspeakBackend
+
+_pl_backend = EspeakBackend(
+    language="pl",
+    preserve_punctuation=True,
+    with_stress=True,
+    words_mismatch="ignore",
+    language_switch="remove-flags",
+)
+
+_PL_WORD = "0-9a-ząćęłńóśźż"
+
+
+def _abbr(k: str, v: str):
+    """Regex dla skrótu: musi stać na granicy słowa. 'r.'/'w.' tylko po liczbie/liczbie rzymskiej
+    (inaczej 'Piotr.' -> 'Piotroku')."""
+    if k in ("r.", "w."):
+        return re.compile(rf"(?<=[0-9ivx])\s?{re.escape(k)}"), f" {v}"
+    tail = "" if k.endswith(".") else rf"(?![{_PL_WORD}])"
+    return re.compile(rf"(?<![{_PL_WORD}]){re.escape(k)}{tail}"), v
+
+
+_PL_ABBREV = [
+    _abbr(k, v)
+    for k, v in [
+        ("np.", "na przykład"),
+        ("tzn.", "to znaczy"),
+        ("tj.", "to jest"),
+        ("itd.", "i tak dalej"),
+        ("itp.", "i tym podobne"),
+        ("m.in.", "między innymi"),
+        ("dr", "doktor"),
+        ("prof.", "profesor"),
+        ("inż.", "inżynier"),
+        ("mgr", "magister"),
+        ("ul.", "ulica"),
+        ("al.", "aleja"),
+        ("r.", "roku"),
+        ("w.", "wieku"),
+        ("godz.", "godzina"),
+        ("min.", "minut"),
+        ("św.", "święty"),
+        ("nr", "numer"),
+        ("tel.", "telefon"),
+        ("zł", "złotych"),
+    ]
+]
+
+
+def _expand_pl_abbreviations(text: str) -> str:
+    for rx, rep in _PL_ABBREV:
+        text = rx.sub(rep, text)
+    return text
+
+
+def _collapse_whitespace(text: str) -> str:
+    return re.sub(r"\s+", " ", text).strip()
+
+
+def _normalize_punct(text: str) -> str:
+    text = re.sub(r"\s*(?:—|–| - )\s*", ", ", text).replace("…", "...")
+    text = text.replace("„", "").replace("”", "").replace("“", "").replace('"', "").replace("«", "").replace("»", "")
+    return text
+
+
+def polish_cleaners(text: str) -> str:
+    """Tekst PL -> IPA (espeak-ng). Liczby zostaw espeakowi – czyta je poprawnie po polsku
+    w mianowniku; jeśli potrzebujesz odmiany, zamień je wcześniej w prepare_dataset."""
+    text = text.lower()
+    text = _normalize_punct(text)
+    text = _expand_pl_abbreviations(text)
+    phonemes = _pl_backend.phonemize([text], strip=True, njobs=1)[0]
+    return _collapse_whitespace(phonemes)

english/polish_mixed_cleaners.py

@@ -0,0 +1,62 @@
+
+
+# --- tts-pl: Polish text with English words (appended by patch_matcha.py after polish_cleaners) ---------------------
+# English words are marked in braces: "Dzięki za {feedback}, wrzucę to na {Slack}a." The braced part is phonemized by
+# espeak-ng en-us, the rest by the Polish pipeline above; a Polish ending glued to a brace ("{Slack}a") is joined to the
+# English phonemes without a space. All en-us phones are in the symbol table (checked: voxcpm/pilot_phonemes.py).
+_en_backend = EspeakBackend(
+    language="en-us",
+    preserve_punctuation=True,
+    with_stress=True,
+    words_mismatch="ignore",
+    language_switch="remove-flags",
+)
+_EN_SPAN = re.compile(r"\{([^{}]+)\}")
+
+
+def _pl_phon(text: str) -> str:
+    text = _expand_pl_abbreviations(_normalize_punct(text.lower()))
+    return _pl_backend.phonemize([text], strip=True, njobs=1)[0] if text.strip() else ""
+
+
+_VOICELESS = ("p", "t", "k", "f", "s", "ʃ", "θ", "h", "tʃ", "ˈp", "ˈt", "ˈk", "ˈf", "ˈs", "ˈʃ", "ˈθ", "ˈh", "ˌp", "ˌt", "ˌk", "ˌf", "ˌs", "ˌh")
+
+
+def _one_letter(w: str, next_ph: str) -> str:
+    voiceless = next_ph.startswith(_VOICELESS)
+    return {"w": "f" if voiceless else "v", "z": "s" if voiceless else "z", "i": "i", "a": "a", "o": "ɔ", "u": "u"}[w]
+
+
+def polish_mixed_cleaners(text: str) -> str:
+    """Polish IPA with English IPA for {braced} spans. Without braces identical to polish_cleaners."""
+    if "{" not in text:
+        return polish_cleaners(text)
+    out, pos = "", 0
+    for m in _EN_SPAN.finditer(text):
+        pl = text[pos:m.start()]
+        prep = ""
+        mm = re.search(r"(?:^|\s)([wzaiouWZAIOU])\s*$", pl)          # a one-letter word right before the English span: espeak
+        if mm:                                                 # would read it alone as a letter name ("w" -> "vɨ")
+            prep, pl = mm.group(1).lower(), pl[:mm.start(1)]
+        if pl.strip():
+            glue = not pl[0].isspace() and out != ""            # Polish ending right after the previous brace
+            ph = _pl_phon(pl)
+            if glue:                                             # the ending is unstressed: part of the English word
+                head, _, rest = ph.partition(" ")
+                ph = head.replace("ˈ", "").replace("ˌ", "") + (" " + rest if rest else "")
+            out += ("" if glue else " ") + ph
+        en = _en_backend.phonemize([m.group(1)], strip=True, njobs=1)[0]
+        if prep:                                                 # "w {weekend}": the preposition, voiced or not by the next sound
+            out += " " + _one_letter(prep, en)
+        glue = m.start() > 0 and not text[m.start() - 1].isspace() and out != ""
+        out += ("" if glue else " ") + en
+        pos = m.end()
+    tail = text[pos:]
+    if tail.strip():
+        glue = not tail[0].isspace()
+        ph = _pl_phon(tail)
+        if glue:
+            head, _, rest = ph.partition(" ")
+            ph = head.replace("ˈ", "").replace("ˌ", "") + (" " + rest if rest else "")
+        out += ("" if glue else " ") + ph
+    return _collapse_whitespace(out)

model/matcha_pl_v2.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c4e8605cf517f932d8f72fc32f4d81854de33c2a7d6fa82ad052fc755b3a024
+size 83542710

onnx/matcha_pl_t2.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2a2ea8bc43fa60ec731299e10dda57459c7375c4f5d40d6506f0514fec2884a
+size 139965159

onnx/matcha_pl_t4.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dabac7caaea9d75690d12bb878cd3dc6163484398ef2610d516df413abc8703
+size 140422993

onnx/voices.json

@@ -0,0 +1 @@
+{"default": "fav2", "targets": [], "presets": {"fav2": {"label": "fav2 (synthetic voice)", "weights": {"0": 1.0}}}, "speakers": {"0": {"name": "fav2", "emb": [-0.171126, 0.322866, -0.493196, -0.251865, -1.496564, -0.773552, 0.672265, -0.925804, -0.569764, -0.062601, 0.368567, -0.555231, -0.010228, 0.525713, 1.064089, 1.084695, -0.108082, -1.248569, 0.401326, -0.809678, 0.146417, 0.637237, 0.170811, -0.746055, 0.30452, -0.758564, -0.460776, -0.647524, -0.687801, 2.077753, -0.159227, -0.075347, 0.608661, 0.566903, -0.023465, -1.022551, -0.390619, -0.467591, 0.456134, 0.381035, -0.043261, -0.79233, -1.292068, -0.338539, 0.090066, -0.007928, 1.320292, -0.026984, -0.945175, -0.495162, -0.239323, 0.335876, -0.208299, 1.380403, -0.118773, -1.187929, 1.033132, 0.574047, 0.126154, 0.615193, -0.638149, -0.652656, 0.079094, 1.707609]}}, "styles": {"0": {"label": "flat-range · slow", "emb": [-0.145681, 0.028514, 0.031073, 0.213126, 0.141315, -0.091171, 0.020456, -0.050942, 0.116065, -0.037322, -0.050199, -0.128892, -0.080357, 0.045101, 0.012145, -0.013334, 0.032367, 0.00523, 0.016566, -0.002529, 0.094458, 0.054667, -7e-06, -0.011532, -0.006527, -0.130739, -0.04346, -0.025447, -0.056555, 0.02478, 0.035768, -0.007505, -0.064534, 0.114622, -0.005323, -0.011425, -0.081168, -0.026979, -0.054348, 0.063037, 0.021073, 0.113498, 0.125257, 0.107236, -0.032194, 0.021723, 0.069703, 0.023842, 0.102836, -0.123746, 0.012705, -0.015574, -0.023926, 0.049007, 0.077019, 0.142015, -0.022084, -0.036438, 0.126887, -0.074276, -0.087569, -0.009835, -0.031884, 0.1038]}, "1": {"label": "flat-range · slow · question", "emb": [-0.065793, 0.16722, -0.040421, 0.10364, 0.149176, -0.021905, -0.0432, -0.050239, -0.040979, -0.090244, -0.097514, -0.184097, -0.077265, 0.04736, -0.055627, -0.118585, 0.066956, -0.061624, 0.048341, 0.052993, 0.177459, 0.01611, -0.109275, -0.070324, 0.041396, 0.006175, 0.020192, -0.063335, 0.00341, -0.05934, -0.201707, 0.057273, 0.006712, 0.016891, 0.020629, -0.063877, -0.011998, -0.184993, -0.020477, 0.005055, -0.082237, 0.120278, -0.023095, 0.012626, 0.12684, 0.096845, -0.126149, 0.069004, -0.104394, 0.008861, 0.031055, -0.069648, 0.03642, 0.04356, 0.024144, 0.107632, 0.092163, 0.005802, 0.17045, -0.057356, -0.10544, 0.03624, -0.143395, 0.086159]}, "2": {"label": "flat-range · normal", "emb": [0.015198, 0.005387, -0.070077, 0.085694, 0.052682, -0.069297, -0.052193, -0.011528, 0.103583, 0.036779, -0.018627, -0.018627, -0.054331, -0.057561, 0.073174, 0.013283, 0.107597, 0.027577, 0.007558, 0.048043, -0.014568, 0.021438, -0.040362, -0.023572, 0.020275, -0.050076, -0.064918, 0.025708, -0.020152, -0.007477, 0.011547, 0.020923, -0.030765, -0.026945, -0.108283, 0.071716, -0.079583, -0.020836, 0.00641, 0.041376, -0.011752, 0.124817, 0.074644, 0.055393, 0.074216, 0.001747, 0.055778, -0.037595, 0.018188, -0.117185, -0.054989, 0.057521, 0.038923, 3.1e-05, -0.018487, 0.037477, 0.0652, 0.003342, 0.0622, 0.02653, 0.015227, -0.062551, 0.058737, -0.038891]}, "3": {"label": "flat-range · normal · question", "emb": [-0.033626, 0.123205, -0.104265, 0.116839, 0.158891, 0.03783, -0.124739, -0.13248, 0.047867, 0.023742, 0.102209, -0.103134, -0.015378, 0.050243, 0.184138, -0.05144, 0.137179, 0.063629, 0.10753, -0.026578, -0.051683, 0.090953, 0.096812, -0.112768, 0.075321, -0.093111, 0.098713, -0.018047, 0.111272, -0.088065, -0.118353, 0.12843, -0.041105, -0.091738, -0.013021, 0.096103, -0.080562, -0.139384, -0.049917, -0.024333, -0.139616, 0.064256, 0.010544, -0.032663, -0.035371, 0.009209, -0.197999, 0.041773, -0.06381, -0.09734, 0.00059, 0.013481, 0.011969, -0.06911, 0.026348, -0.020739, 0.100743, 0.028176, 0.141968, 0.027852, -0.026274, 0.098182, -0.024583, 0.015864]}, "4": {"label": "flat-range · fast", "emb": [0.079858, -0.032991, -0.141646, -0.0183, -0.05767, -0.010497, -0.130122, -0.063902, 0.006765, 0.108658, 0.042958, -0.02097, 0.050789, -0.10057, 0.101453, 0.113868, 0.042049, 0.099299, -0.084551, -0.002367, -0.086793, 0.08406, -0.099978, 0.061491, 0.097875, 0.108743, -0.111331, 0.084997, 0.144157, 0.00984, 0.065044, 0.041029, 0.064102, -0.095855, -0.145943, 0.078539, 0.045341, 0.036196, 0.068478, -0.028343, -0.042336, 0.131945, 0.046076, -0.006211, 0.078042, -0.122365, 0.007158, -0.175239, -0.052579, -0.182534, -0.110177, 0.126633, 0.037948, 0.037947, -0.06254, -0.004245, 0.158694, 0.066889, -0.018682, -0.003524, 0.125085, 0.024064, 0.130969, -0.106359]}, "5": {"label": "flat-range · fast · question", "emb": [0.09611, 0.034653, -0.220507, -0.148164, 0.024787, -0.047649, -0.211915, 0.173699, -0.072696, 0.068309, 0.094034, 0.072822, 0.019759, -0.108546, 0.158991, -0.032414, -0.033625, 0.035397, -0.038269, 0.020347, -0.026651, 0.06203, -0.057607, -0.056806, 0.105391, 0.188259, 0.159449, 0.088129, 0.173849, 0.078067, -0.155068, 0.117547, 0.10826, -0.139548, -0.019679, 0.050196, -0.005208, -0.135763, -0.048872, -0.156271, -0.068358, 0.079436, -0.067806, -0.098738, -0.000494, -0.008177, -0.207378, -0.158861, -0.088763, -0.091702, -0.054759, 0.171019, 0.123302, -0.038804, -0.038359, -0.009134, 0.188344, 0.213565, 0.065066, 0.131621, 0.000913, 0.027884, 0.028019, -0.047102]}, "6": {"label": "mid-range · slow", "emb": [-0.108414, -0.003766, 0.026759, 0.090972, 0.10488, 0.021303, 0.077811, 0.025143, 0.086748, -0.043222, -0.055501, -0.021596, -0.058954, 0.001334, 0.023127, -0.064402, -0.036275, -0.027464, -0.026106, -0.042851, 0.060475, -0.092045, 0.021875, 0.021948, -0.03574, -0.059152, -0.073984, 0.022593, -0.105647, 0.0545, 0.035674, -0.054579, -0.073088, 0.100104, 0.035317, -0.017922, -0.046948, -0.142279, -0.01218, 0.065619, 0.045547, -0.026938, 0.060306, 0.052985, 0.078487, 0.10227, 0.052383, 0.094831, 0.118608, 0.057531, 0.049244, -0.069276, 0.011932, -0.007232, 0.115472, 0.066428, -0.100617, -0.001337, 0.06223, -0.00322, -0.097148, -0.034642, -0.11133, 0.029218]}, "7": {"label": "mid-range · slow · question", "emb": [-0.029557, 0.133307, 0.098692, 0.122858, 0.09645, 0.008557, -0.014066, -0.135018, -0.079446, -0.12148, -0.135168, -0.129205, -0.064973, 0.125791, -0.074336, 0.01901, -0.015607, -0.046001, 0.083561, -0.01328, 0.075397, -0.012248, 0.024806, -0.042582, -0.084804, -0.042625, 0.019341, -0.129714, 0.029283, 0.029565, -0.080985, 0.070864, -0.059116, 0.07597, 0.106906, -0.044583, -0.017669, -0.20186, -0.141748, 0.010677, 0.008098, -0.023605, 0.056758, 0.098303, -0.034178, 0.045052, -0.183866, 0.117161, -0.002713, -0.024853, 0.015352, -0.082478, -0.014168, 0.110169, -0.000971, -0.01839, 0.054924, -0.169824, 0.123016, -0.045664, -0.173564, 0.084706, -0.13797, -0.031571]}, "8": {"label": "mid-range · normal", "emb": [0.047679, -0.033757, 0.012959, -0.020053, -0.017468, -0.00887, 0.010025, 0.062036, 0.099756, 0.013468, -0.017102, -0.037922, 0.03251, 0.001108, 0.008804, 0.02545, -9e-05, -0.001319, -0.010407, 0.004526, -0.03449, -0.04883, 0.001763, 0.010887, -0.014453, 0.03218, -0.055419, 0.059566, -0.003377, 0.022355, 0.007696, -0.022474, 0.013298, 0.010364, -0.023911, 0.041498, -0.037128, 0.003917, 0.013321, -0.010775, -0.001142, -0.059385, 0.010826, 0.003236, 0.048636, 0.028829, 0.040092, 0.028113, 0.009203, -0.027532, 0.003704, 0.009829, 0.002783, -0.042211, 0.032808, 0.036177, -0.001528, 0.017659, -0.00515, 0.016582, 0.046999, -0.007643, 0.016614, -0.050302]}, "9": {"label": "mid-range · normal · question", "emb": [0.100659, -0.035835, -0.122456, -0.055806, -0.030154, 0.026734, -0.065558, -0.071907, -0.035, -0.033529, -0.074552, -0.03608, -0.054054, -0.077936, -0.061722, -0.055407, 0.160748, 0.000847, -0.061431, 0.010338, 0.100741, 0.014006, -0.102683, -0.072087, 0.03347, 0.006653, 0.183114, -0.049406, -0.037271, 0.041016, -0.186704, -0.042957, 0.027818, 0.038135, 0.052061, 0.089554, 0.044856, -0.021578, -0.071574, 0.037154, -0.032053, -0.05052, -0.023911, -0.050673, -0.013418, 0.040641, -0.213389, 0.059986, -0.13572, 0.073084, -0.129966, -0.05335, 0.188839, 0.034079, -0.022861, -0.074348, 0.077725, -0.061973, -0.01645, 0.041444, -0.007721, 0.011829, -0.024506, 0.00019]}, "10": {"label": "mid-range · fast", "emb": [0.148559, -0.117011, -0.108052, -0.079113, -0.151191, -0.009634, -0.05918, 0.006944, 0.0017, 0.115003, 0.033147, 0.08128, 0.118907, -0.075013, -0.014178, 0.08572, -0.014366, 0.092867, -0.012872, -0.028587, -0.077081, 0.010422, -0.063002, 0.000577, -0.009214, 0.135065, -0.056282, 0.013967, 0.067966, -0.058367, 0.023978, 0.070172, 0.075121, -0.092142, -0.06845, 0.120971, 0.014942, 0.059432, 0.032089, -0.000947, -0.023865, -0.014222, 0.013042, -0.099835, -0.027346, -0.049702, 0.015537, -0.066083, -0.075356, -0.094995, -0.060072, 0.054041, 0.036734, -0.039968, -0.078773, -0.020267, -0.01236, 0.055531, -0.081482, 0.057737, 0.08034, -0.001431, 0.14718, -0.100518]}, "11": {"label": "mid-range · fast · question", "emb": [0.073619, 0.043859, -0.05577, -0.034364, -0.257909, 0.070455, -0.16292, -0.010065, -0.044395, 0.102808, -0.050129, 0.0522, 0.193897, -0.029656, 0.007619, 0.052111, 0.04524, 0.049192, 0.113122, 0.129318, -0.029497, 0.10029, -0.2246, 0.008363, 0.124819, 0.078415, 0.031787, -0.101723, 0.063401, -0.14017, -0.119636, 0.044516, 0.090903, -0.005523, -0.089667, 0.139016, 0.156393, 0.024767, -0.054849, -0.176828, -0.139236, -0.009136, -0.24576, -0.182452, -0.160701, -0.003506, -0.06909, -0.035963, -0.067486, -0.004653, 0.015554, 0.119868, 0.013078, -0.071847, -0.064107, -0.047181, 0.090308, 0.078807, 0.023879, 0.118498, 0.09614, 0.184368, -0.05321, -0.100499]}, "12": {"label": "wide-range · slow", "emb": [-0.11099, 0.018829, 0.151204, 0.0139, 0.021509, 0.077328, 0.178788, -0.00928, 0.037616, -0.079435, -0.026652, 0.004519, -0.03194, -0.02984, -0.083479, 0.028554, -0.129474, -0.073987, 0.002037, -0.035017, 0.00193, -0.117057, 0.072726, -0.015901, -0.111475, -0.012635, -0.029585, -0.043885, -0.097188, -0.00683, 0.040953, -0.091716, -0.097066, 0.125877, 0.101466, -0.117703, -0.001483, 0.087936, -0.00806, 0.008189, 0.058145, -0.089602, 0.007832, 0.081087, -0.011696, 0.06454, 0.071723, 0.129059, 0.07864, 0.163052, 0.116826, -0.039481, -0.059053, -0.109745, 0.080791, 0.026902, -0.097094, -0.036005, -0.098446, 0.011695, -0.115028, 0.013473, -0.030081, 0.050539]}, "13": {"label": "wide-range · slow · question", "emb": [-0.049316, 0.079729, 0.134762, -0.017171, 0.016638, 0.03792, 0.077599, -0.047527, -0.058096, -0.125282, -0.065962, 0.054827, -0.042702, 0.080202, -0.128793, 0.047482, -0.015366, -0.128568, 0.174576, -0.064135, 0.028825, -0.057269, 0.013506, -0.084323, -0.018376, -0.006145, 0.135988, -0.051993, 0.045764, -0.068694, -0.096576, 0.060418, -0.033247, 0.118162, 0.214827, -0.079408, 0.013513, -0.143196, -0.043047, -0.070651, 0.008669, -0.048957, -0.056213, 0.003604, -0.062643, 0.060966, -0.091224, 0.131408, -0.051059, 0.116348, 0.082638, -0.183355, 0.106217, 0.073501, 0.069447, 0.046283, 0.073742, -0.094445, -0.096596, 0.057557, -0.018585, 0.11509, -0.091432, 0.025908]}, "14": {"label": "wide-range · normal", "emb": [0.001553, -0.012023, 0.022358, -0.114422, -0.004232, 0.085474, 0.091031, -0.031596, 0.057701, -0.040212, -0.030307, 0.059041, 0.029353, -0.018229, -0.066481, 0.040342, -0.084407, -0.071293, -0.011668, -0.019989, -0.097087, -0.081197, 0.043118, 0.05649, -0.095185, 0.082748, 0.037413, 0.053604, 0.048286, 0.028822, -0.014488, -0.034348, 0.006524, -0.035016, 0.036618, 0.001304, -0.009479, 0.160705, -0.002441, 0.008384, 0.018137, -0.126345, -0.07473, -0.048525, -0.066345, -0.034876, 0.039267, 0.043862, 0.02485, 0.089538, 0.044936, -0.034978, 0.019987, -0.038721, -0.009296, 0.00596, -0.083425, -0.002204, -0.127395, 0.020764, -0.041537, 0.060425, 0.03617, -0.042043]}, "15": {"label": "wide-range · normal · question", "emb": [0.031989, 0.12776, 0.097385, -0.115555, 0.019236, 0.075017, 0.006442, -0.068154, -0.075861, -0.122641, -0.011926, 0.1815, -0.024032, -0.059157, -0.206298, 0.076823, -0.15704, -0.082074, 0.253635, 0.075241, 0.020136, -0.032513, -0.070692, -0.050269, -0.057442, 0.141685, 0.191198, -0.020758, 0.109362, -0.140957, -0.093134, 0.079722, 0.104948, -0.135023, 0.117103, -0.141334, -0.067662, -0.041133, -0.069584, -0.123385, -0.084509, -0.171221, -0.072137, -0.091094, -0.012565, 0.058296, -0.145974, 0.03443, -0.038373, 0.012998, 0.006425, -0.059752, 0.043147, -0.073827, 0.020636, 0.113688, 0.142444, -0.03517, 0.023351, 0.126022, -0.010433, 0.027106, 0.032558, 0.033757]}, "16": {"label": "wide-range · fast", "emb": [0.115599, -0.161136, -0.017431, -0.194831, -0.190049, 0.115864, -0.027475, -0.059261, 0.004603, 0.053946, 0.058454, 0.123359, 0.167072, -0.035785, -0.042583, 0.098873, -0.055196, -0.016422, -0.073254, -0.057121, -0.117421, -0.035423, -0.049022, 0.022217, -0.045531, 0.14901, 0.03188, 0.039933, 0.114307, -0.010411, 0.067121, 0.07167, 0.077926, -0.039093, 0.018199, 0.13719, 0.112801, 0.193481, 0.011642, -0.087589, 0.05298, -0.081547, -0.037497, -0.120967, -0.089402, -0.108299, 0.098022, -0.054007, -0.052677, 0.008059, 0.001495, 0.017505, -0.021015, -0.057143, -0.132628, -0.041625, 0.020005, 0.046707, -0.255037, 0.038068, 0.124554, 0.037912, 0.14988, -0.066634]}, "17": {"label": "wide-range · fast · question", "emb": [0.144282, -0.070793, -0.002914, -0.1106, -0.186188, 0.1517, -0.046887, -0.091894, -0.099903, 0.008947, 0.039248, 0.194223, 0.059878, -0.129902, 0.013289, -0.02141, 0.007187, 0.010225, 0.133719, -0.010978, -0.103538, 0.001154, 0.001617, 0.02783, 0.174246, 0.161041, 0.069097, -0.014994, 0.140037, -0.069386, -0.149788, -0.034438, 0.068539, -0.06458, -0.006801, -0.000357, 0.067643, -0.222343, -0.03566, -0.223449, -0.043558, -0.130799, -0.064022, -0.154887, -0.037838, -0.038297, -0.081953, 0.020593, -0.169838, 0.140732, -0.054224, 0.050515, 0.021453, -0.180905, -0.079634, -0.081443, 0.068684, 0.10073, -0.081147, 0.084175, 0.024963, 0.050901, 0.029197, -0.016361]}}, "formula": "spk_emb = speakers['0'].emb (+ styles[k].emb for a style token; 7 = neutral)", "defaults": {"temperature": 0.8, "length_scale": 0.8, "style": 7}}

samples/README.md

@@ -0,0 +1,16 @@
+Held-out test sentences (not in the training set). v2 at the default settings (tempo 0.8, temperature 0.8, 4 ODE steps); teacher = VoxCPM2 with the frozen fav2 voice.
+
+| # | text | v2 | VoxCPM2 teacher |
+|---|---|---|---|
+| e01 | Dzień dobry! W czym mogę dziś pomóc? | [v2](samples/e01_v2.wav) | [teacher](samples/e01_teacher.wav) |
+| e02 | Proszę zamknąć okno, robi się zimno. | [v2](samples/e02_v2.wav) | [teacher](samples/e02_teacher.wav) |
+| e03 | Czy mam włączyć światło w kuchni? | [v2](samples/e03_v2.wav) | [teacher](samples/e03_teacher.wav) |
+| e04 | Kiedy przyjedzie kurier z paczką? | [v2](samples/e04_v2.wav) | [teacher](samples/e04_teacher.wav) |
+| e05 | Spotkanie zaczyna się o piętnastej trzydzieści, a potrwa mniej więcej godzinę. | [v2](samples/e05_v2.wav) | [teacher](samples/e05_teacher.wav) |
+| e06 | Po meetingu wyślę ci feedback na maila, a w weekend zrobię update na laptopie. | [v2](samples/e06_v2.wav) | [teacher](samples/e06_teacher.wav) |
+| e07 | Sprawdziłem pogodę: jutro rano będzie dwanaście stopni i lekki deszcz, więc lepiej weź parasol. | [v2](samples/e07_v2.wav) | [teacher](samples/e07_teacher.wav) |
+| e08 | Tak. | [v2](samples/e08_v2.wav) | [teacher](samples/e08_teacher.wav) |
+| e09 | Uwaga, podłoga jest mokra! Proszę iść ostrożnie. | [v2](samples/e09_v2.wav) | [teacher](samples/e09_teacher.wav) |
+| e10 | Zanim zaczniemy gotować, umyjmy ręce. Potem pokroimy warzywa w kostkę i wrzucimy je na rozgrzaną patelnię. Na koniec dodamy odrobinę soli i pieprzu. Gotowe w dwadzieścia minut! | [v2](samples/e10_v2.wav) | [teacher](samples/e10_teacher.wav) |
+| e11 | Wczoraj wieczorem wybrałem się na krótki spacer nad rzekę. Było cicho, tylko kaczki pływały przy brzegu, a niebo powoli robiło się pomarańczowe. Pomyślałem wtedy, że takie chwile są naprawdę bezcenne. A ty, jak lubisz spędzać wieczory? | [v2](samples/e11_v2.wav) | [teacher](samples/e11_teacher.wav) |
+| e12 | Nazywam się HVSR i zostałem zbudowany przez firmę Machinekind, żeby pomagać ludziom w codziennych sprawach. | [v2](samples/e12_v2.wav) | [teacher](samples/e12_teacher.wav) |

samples/e01_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1399e8fae07a4bf43beba64bc9e95cd66e7136f854947a0bbe47e867c221cd0a
+size 103532

samples/e05_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da49be5db5fffe6b0337d814318930a61b70cf8ef4521d94d5e317c27e44ff5b
+size 188910

samples/e05_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47ceb30103ffabee549e0d56dc3339b6d819f78ca925d07833a33cf747c6e11b
+size 183852

samples/e06_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:526375bab41ff55a5d22e43056c2a179d270df20d16ad3b372be55ff1b3115a1
+size 191498

samples/e06_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70eab2acde73c51e6f9ecc2b05da412686c2d7ad770aa38fcc2e804f6c39e465
+size 183852

samples/e07_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3413e8cfcc33dbf71b51ed858ed9d90291217f3410da0527c36c269287f6ad04
+size 246300

samples/e07_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d13152a9fa679f813a1ef62e76305a273dc6f21704937b87b689b49232ea466
+size 236076

samples/e09_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae665cf8bdc7f62ec575fb7307d55ce02d38203c70d9933ee90b251a97a14d06
+size 143516

samples/e09_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7818a4793c2820af4f46114203c84ae74260825e1a8fc42eac1dcbe62d2983a5
+size 129068

samples/e10_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bc1855a6fcb110e983b675394f9d4e99192c6c54f461e4992a32dd0a9b0dcea
+size 467388

samples/e10_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2e02b8f60c2637fc5ad2f738de938eb2e4d483f28bdae0d9bf3819e835d0617
+size 442924

samples/e11_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c49d533bd0953b808495f0d54992c4851f79e97deccde9ea9f2ba004483a7d43
+size 580284

samples/e11_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:205fe76d82dfcf7c526e20f0246155e2ee7286066aabcc52be9f5be62b17681b
+size 573484

samples/e12_teacher.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:756a23ca8236efae553edd8865d1023b1607cdd3f1b4a84dc5f9d407261ba068
+size 308628

samples/e12_v2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:068d1ff1ba93579476d93c648a2d7b156be2cda35cd90d6481b7dc6747b0abb7
+size 269868

vocoder/config.json

@@ -0,0 +1,62 @@
+{
+  "resblock": "1",
+  "num_gpus": 0,
+  "batch_size": 16,
+  "learning_rate": 2e-05,
+  "adam_b1": 0.8,
+  "adam_b2": 0.99,
+  "lr_decay": 0.999,
+  "seed": 1234,
+  "upsample_rates": [
+    8,
+    8,
+    2,
+    2
+  ],
+  "upsample_kernel_sizes": [
+    16,
+    16,
+    4,
+    4
+  ],
+  "upsample_initial_channel": 512,
+  "resblock_kernel_sizes": [
+    3,
+    7,
+    11
+  ],
+  "resblock_dilation_sizes": [
+    [
+      1,
+      3,
+      5
+    ],
+    [
+      1,
+      3,
+      5
+    ],
+    [
+      1,
+      3,
+      5
+    ]
+  ],
+  "segment_size": 8192,
+  "num_mels": 80,
+  "num_freq": 1025,
+  "n_fft": 1024,
+  "hop_size": 256,
+  "win_size": 1024,
+  "sampling_rate": 22050,
+  "fmin": 0,
+  "fmax": 8000,
+  "fmax_for_loss": null,
+  "num_workers": 8,
+  "dist_config": {
+    "dist_backend": "nccl",
+    "dist_url": "tcp://localhost:54321",
+    "world_size": 1
+  },
+  "fine_tuning": true
+}

vocoder/generator_fav2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd4abd345bed0a61fd67a457b8ee8c5353b8c43270ad94a58efa8ff41cd1912b
+size 55824872