Stage verified Inflect v2 release candidate and evaluation evidence

README.md

@@ -3,218 +3,269 @@ license: apache-2.0
 language:
   - en
 pipeline_tag: text-to-speech
+model_name: Inflect-Micro-v2
+metrics:
+  - wer
 tags:
   - text-to-speech
   - speech-synthesis
   - local-tts
   - cpu
+  - edge-ai
   - small-model
+  - pytorch
   - vits
   - 24khz
-library_name: pytorch
+thumbnail: assets/inflect-v2-repository-hero.png
+inference: false
 ---
 
 ![Inflect v2](assets/inflect-v2-repository-hero.png)
 
 <h1 align="center">Inflect-Micro-v2</h1>
-<p align="center"><strong>Complete local TTS under 10M parameters.</strong><br>A complete 24 kHz English text-to-waveform model with its neural waveform decoder included.</p>
-<p align="center"><a href="https://huggingface.co/spaces/owensong/Inflect-v2-Private-Playground"><strong>Try it in the browser</strong></a> · <a href="https://huggingface.co/owensong/Inflect-Nano-v2">Inflect-Nano-v2</a> · <a href="#evaluation">Evaluation</a> · <a href="#limitations">Limitations</a></p>
+<p align="center"><strong>Complete local text-to-waveform speech synthesis under 10M parameters.</strong><br>
+24 kHz English TTS with the neural waveform decoder already inside the model.</p>
+
+<p align="center">
+  <a href="https://huggingface.co/spaces/owensong/Inflect-v2"><img alt="Live demo" src="https://img.shields.io/badge/Live_demo-1769E0?style=for-the-badge&logo=huggingface&logoColor=white"></a>
+  <a href="https://huggingface.co/owensong/Inflect-Nano-v2"><img alt="Inflect-Nano-v2" src="https://img.shields.io/badge/Inflect--Nano--v2-FFFFFF?style=for-the-badge&logo=huggingface&logoColor=1769E0"></a>
+  <a href="https://github.com/owenawsong/Inflect"><img alt="GitHub" src="https://img.shields.io/badge/GitHub-10253F?style=for-the-badge&logo=github&logoColor=white"></a>
+</p>
+
+<p align="center">
+  <a href="#run-locally">Run locally</a> ·
+  <a href="#evaluation">Evaluation</a> ·
+  <a href="docs/API.md">API</a> ·
+  <a href="docs/DEPLOYMENT.md">Deployment</a> ·
+  <a href="docs/EVALUATION.md">Raw protocol</a> ·
+  <a href="https://huggingface.co/collections/owensong/inflect-v2-6a619e820808eccf361a2948">Inflect v2 collection</a>
+</p>
+
+<p align="center"><strong>9.36M complete parameters</strong> · <strong>37.53 MB FP32 weights</strong> · <strong>one fixed English voice</strong> · <strong>no external vocoder</strong></p>
 
-Inflect-Micro-v2 is the quality-focused member of Inflect v2. The complete deployable system has **9,356,513 parameters**; there is no second vocoder download and no hidden large language model at inference time.
+---
 
-## Listen First
+## Listen
 
-Every clip below uses held-out text rather than a training-audio reconstruction.
+These are held-out text generations, not reconstructions of training audio. Each transcript is shown exactly as passed to the public frontend.
 
-| Prompt type | Audio |
-| --- | --- |
-| Conversational | <audio controls preload="none" src="samples/male/conversational.wav"></audio> |
-| Punctuation | <audio controls preload="none" src="samples/male/punctuation.wav"></audio> |
-| Numbers | <audio controls preload="none" src="samples/male/numbers.wav"></audio> |
-| Names and places | <audio controls preload="none" src="samples/male/names_places.wav"></audio> |
-| Technical | <audio controls preload="none" src="samples/male/technical.wav"></audio> |
+| Test | Exact transcript | Generated audio |
+| --- | --- | --- |
+| **Conversational** | It wasn't until later that I realized what had actually happened. | <audio controls preload="metadata" src="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/conversational.wav"></audio><br><a href="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/conversational.wav">download WAV</a> |
+| **Punctuation** | First, close the window; second, turn off the lamp; finally, lock the door. | <audio controls preload="metadata" src="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/punctuation.wav"></audio><br><a href="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/punctuation.wav">download WAV</a> |
+| **Numbers** | The package weighs twelve point six kilograms and arrived on July twenty-first. | <audio controls preload="metadata" src="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/numbers.wav"></audio><br><a href="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/numbers.wav">download WAV</a> |
+| **Names and places** | Gwendolyn photographed the eucalyptus trees outside Ljubljana. | <audio controls preload="metadata" src="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/names_places.wav"></audio><br><a href="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/names_places.wav">download WAV</a> |
+| **Technical** | The system runs on three core components that all have to stay in sync. | <audio controls preload="metadata" src="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/technical.wav"></audio><br><a href="https://huggingface.co/owensong/Inflect-Micro-v2/resolve/main/samples/male/technical.wav">download WAV</a> |
 
-## At a Glance
+## Choose the right Inflect
 
-| | Inflect-Micro-v2 |
-| --- | ---: |
-| Complete deployable parameters | **9,356,513** |
-| FP32 model weights | **37.53 MB** |
-| Sample rate | **24 kHz** |
-| Voice | **1 fixed English male voice** |
-| External vocoder | **None** |
-| Release format | **PyTorch FP32** |
-| Long text | **Punctuation-aware chunking** |
+| | **Inflect-Nano-v2** | **Inflect-Micro-v2** |
+| --- | ---: | ---: |
+| Complete parameters | 3,966,721 | 9,356,513 |
+| FP32 weights | 15.97 MB | 37.53 MB |
+| Positioning | Smallest practical footprint | Strongest Inflect v2 quality |
+| Warm CPU throughput¹ | 1.85× real time | 1.35× real time |
+| Waveform decoder | Included | Included |
+| Public API and frontend | Same | Same |
 
-Inflect v2 is released in two deliberately different sizes. **Micro** spends its larger budget on waveform capacity and hidden width; **Nano** prioritizes footprint while preserving the same public runtime and text frontend. Both numbers above count the complete deployable generator, including the waveform decoder.
+**Inflect-Micro-v2** is the quality-focused member of the family. Both parameter totals count the complete generator, including the 24 kHz neural waveform decoder. There is no hidden inference-time teacher, language model, or second vocoder download.
 
-## Human Listening
+<sub>¹ Host-specific engineering snapshot on an AMD EPYC 7K62. See [Runtime methodology](#runtime-methodology) before comparing hardware.</sub>
 
-![Community blind listening](assets/evidence/human-preference.svg)
+## Run locally
 
-Inflect-Micro-v2 recorded a **66.2% preference rate** (21 wins, 10 losses, 3 ties) in the final anonymous community study. The full study contains **147 pairwise decisions** across the published lineup. Ties count as half a win. This is descriptive community evidence, not a formal MOS study; pair coverage and vote counts differ by system.
+### Install
 
-## Evaluation
+```bash
+git clone https://huggingface.co/owensong/Inflect-Micro-v2
+cd Inflect-Micro-v2
+python -m pip install -r requirements.txt
+```
 
-### Quality versus footprint
+### Python
 
-![Predicted quality versus footprint](assets/evidence/quality-vs-footprint.svg)
+```python
+from inference import InflectTTS
 
-### Three-ASR intelligibility
-
-![Semantic WER across Qwen3-ASR, Nemotron 3.5, and Whisper large-v3](assets/evidence/modern400-three-asr.svg)
+tts = InflectTTS(".", device="cpu")
+tts.save(
+    "A small voice can still have something meaningful to say.",
+    "sample.wav",
+    speed=1.0,
+    variation=0.667,
+    seed=7,
+)
+```
 
-### CPU throughput
+### Download through the Hub
 
-![Warm CPU throughput](assets/evidence/cpu-throughput.svg)
+```python
+import sys
+from huggingface_hub import snapshot_download
 
-### Quality and speed
+model_dir = snapshot_download("owensong/Inflect-Micro-v2")
+sys.path.insert(0, model_dir)
 
-![Quality and CPU throughput](assets/evidence/quality-vs-cpu-speed.svg)
+from inference import InflectTTS
 
-### Complete model footprint
+tts = InflectTTS(model_dir, device="cpu")
+sample_rate, waveform = tts.synthesize("The complete model runs locally.")
+```
 
-![Model footprint](assets/evidence/model-footprint.svg)
+The result is a 24 kHz mono `float32` waveform. Long input is split at punctuation-aware boundaries, synthesized chunk by chunk, and joined with controlled pauses.
 
-| System | Weights | UTMOS22 ↑ | Qwen3 WER ↓ | Nemotron WER ↓ | Whisper WER ↓ | ASR avg. ↓ |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: |
-| **Inflect-Micro-v2** | 37.5 MB | 4.395 | **2.52%** | **5.45%** | **2.73%** | **3.57%** |
-| **Inflect-Nano-v2** | 16.0 MB | 4.386 | **2.79%** | **5.63%** | **2.65%** | **3.69%** |
-| KittenTTS Nano · Bruno | 56.8 MB | 4.217 | 2.15% | 3.96% | 2.17% | 2.76% |
-| KittenTTS Nano · Hugo | 56.8 MB | 4.191 | 2.39% | 3.80% | 2.11% | 2.77% |
-| Piper · Ryan Low | 63.1 MB | 4.289 | 2.81% | 5.51% | 2.87% | 3.73% |
-| Piper · Danny Low | 63.1 MB | 4.195 | 2.62% | 5.60% | 2.55% | 3.59% |
-| Supertonic 3 · M2 · 3-step | 398.1 MB | 2.471 | 3.03% | 6.04% | 3.22% | 4.10% |
-| Supertonic 3 · M2 · 8-step | 398.1 MB | 4.295 | 2.05% | 3.56% | 8.08% | 4.56% |
-
-Inflect-Micro-v2 measured **4.395 UTMOS22** on the matched quality set (95% bootstrap CI 4.381–4.408). On the separately frozen Modern400 intelligibility set it measured **2.52% semantic WER** with Qwen3-ASR, **5.45%** with Nemotron 3.5 ASR, and **2.73%** with Whisper large-v3. Their unweighted corpus-level macro-average is **3.57%**. UTMOS22 is an automated quality predictor, not human MOS; WER measures transcript recoverability, not naturalness.
-
-The corrected Supertonic run uses version 1.3.1, official voice `M2`, clean generations, and explicit 16 kHz ASR input normalization. Whisper still produces insertion-heavy hallucinations on a small subset of otherwise intelligible Supertonic 8-step clips, yielding **8.08%** versus **2.05% / 3.56%** from Qwen and Nemotron. Individual recognizer scores therefore remain visible beside the requested descriptive average.
+## Why Inflect
 
-<details>
-<summary><strong>Protocol and raw evidence</strong></summary>
-
-- 400 identical unseen English prompts per system; 3,200 generated clips total.
-- The frozen set combines 200 fixed modern/stress prompts with 200 deterministic Google FLEURS `en_us` test prompts.
-- Exact-text exclusion was run against 87,362 training transcripts.
-- Semantic WER is reported independently for `Qwen/Qwen3-ASR-1.7B`, `nvidia/nemotron-3.5-asr-streaming-0.6b`, and `openai/whisper-large-v3` with the same disclosed English normalizer.
-- `ASR avg.` is the unweighted mean of the three corpus-level semantic WERs; it is descriptive and never replaces the individual scores.
-- UTMOS22 uses `tarepan/SpeechMOS` v1.2.0.
-- Headline intervals use 10,000 bootstrap samples.
-- Kitten and Piper voices are reported separately; Supertonic 3-step and 8-step are never pooled.
-- The Modern400 corpus hash is `b7504ce2dce44a2da82770a6a5dfd2a034fe17e2113980f8a69663ade417a34c`.
-- Prompts, summary data, and compressed row-level ASR reports are under `evaluation/final/modern400/`.
+| **Small enough to ship** | **Complete by design** |
+| --- | --- |
+| 37.53 MB of FP32 model weights, with no server dependency. | Text frontend, acoustic generator, duration model, and waveform decoder ship together. |
+| **Built for local inference** | **Measured, not hand-picked** |
+| CPU-ready PyTorch runtime, deterministic seeds, and punctuation-aware long text. | Frozen prompts, raw hypotheses, confidence intervals, hashes, and per-system reports are included. |
 
-</details>
+## Evaluation
 
-## Quickstart
+No single metric captures TTS quality. Inflect v2 reports **human preference**, **predicted naturalness**, **three-ASR intelligibility**, **footprint**, and **runtime** separately rather than hiding them inside one composite score.
 
-```bash
-pip install -r requirements.txt
-python inference.py \
-  --model-dir . \
-  --device cpu \
-  --text "A small voice can still have something meaningful to say." \
-  --output sample.wav
-```
+### 1. Human blind preference
 
-```python
-import sys
-from huggingface_hub import snapshot_download
+![Community blind listening](assets/evidence/human-preference.svg)
 
-model_dir = snapshot_download("owensong/Inflect-Micro-v2")
-sys.path.insert(0, model_dir)
-from inference import InflectTTS
+Inflect-Micro-v2 recorded a **66.2% preference rate** (21 wins · 10 losses · 3 ties) in the final anonymous community study. Systems were hidden, left/right order was randomized, and ties count as half a win. This is descriptive community evidence, not formal MOS.
 
-tts = InflectTTS(model_dir, device="cpu")
-tts.save(
-    "The complete speech model runs locally.",
-    "sample.wav",
-    speed=1.0,
-    variation=0.667,
-    seed=7,
-)
-```
+### 2. Predicted naturalness versus footprint
+
+![Predicted quality versus footprint](assets/evidence/quality-vs-footprint.svg)
+
+The 500-prompt UTMOS22 quality run used the same unseen text per system. KittenTTS and Piper are summarized as equal-weight two-voice means in the figure, with the observed voice range visible. Supertonic 3-step is annotated below the plotted quality range rather than compressing the useful portion of the chart.
+
+| System / voice | Weights | UTMOS22 ↑ |
+| --- | ---: | ---: |
+| **Inflect-Micro-v2** | 37.5 MB | **4.395** |
+| **Inflect-Nano-v2** | 16.0 MB | **4.386** |
+| KittenTTS Nano · Bruno | 56.8 MB | 4.217 |
+| KittenTTS Nano · Hugo | 56.8 MB | 4.191 |
+| Piper Low · Ryan | 63.1 MB | 4.289 |
+| Piper Low · Danny | 63.1 MB | 4.195 |
+| Supertonic 3 · James · 8-step | 398.1 MB | 4.295 |
+| Supertonic 3 · James · 3-step | 398.1 MB | 2.471 |
 
-The packaged runtime accepts `speed` from `0.5` to `2.0`, `variation` from `0.0` to `1.0`, and a deterministic integer seed. Long input is split at punctuation-aware boundaries and concatenated with controlled pauses.
+Inflect-Micro-v2: **4.395 UTMOS22**, 95% bootstrap CI **4.381–4.408**. UTMOS22 is a learned predictor, not human MOS.
 
-### Controls
+### 3. Intelligibility across three ASR systems
 
-| Control | Default | Effect |
-| --- | ---: | --- |
-| `speed` | `1.0` | Speech rate; supported range `0.5`–`2.0` |
-| `variation` | `0.667` | Stochastic latent variation; lower is steadier, higher is more variable |
-| `seed` | `0` | Reproducible sampling seed; incremented per long-text chunk |
+![Semantic WER across Qwen3-ASR, Nemotron 3.5, and Whisper large-v3](assets/evidence/modern400-three-asr.svg)
 
-Use a fixed seed when comparing systems. A different seed is a different stochastic sample, not a different checkpoint.
+| System / voice | Qwen3-ASR ↓ | Nemotron 3.5 ↓ | Whisper large-v3 ↓ | Descriptive mean ↓ |
+| --- | ---: | ---: | ---: | ---: |
+| **Inflect-Micro-v2** | **2.52%** | **5.45%** | **2.73%** | **3.57%** |
+| **Inflect-Nano-v2** | **2.79%** | **5.63%** | **2.65%** | **3.69%** |
+| KittenTTS Nano · Bruno | 2.15% | 3.96% | 2.17% | 2.76% |
+| KittenTTS Nano · Hugo | 2.39% | 3.80% | 2.11% | 2.77% |
+| Piper Low · Danny | 2.62% | 5.60% | 2.55% | 3.59% |
+| Piper Low · Ryan | 2.81% | 5.51% | 2.87% | 3.73% |
+| Supertonic 3 · M2 · 3-step | 3.03% | 6.04% | 3.22% | 4.10% |
+| Supertonic 3 · M2 · 8-step | 2.05% | 3.56% | 8.08% | 4.56% |
 
-## Architecture
+Inflect-Micro-v2: **2.52% Qwen3-ASR**, **5.45% Nemotron 3.5**, and **2.73% Whisper large-v3** semantic WER. The **3.57%** mean is descriptive only; individual recognizers remain visible because Whisper produced insertion-heavy hallucinations on a small subset of otherwise intelligible Supertonic 8-step clips.
 
-Inflect v2 is a **parameter-efficient VITS-family, end-to-end text-to-waveform speech generator** with a phoneme frontend, monotonic alignment, stochastic latent synthesis, and an integrated neural waveform decoder.
+<details>
+<summary><strong>Open the frozen evaluation protocol</strong></summary>
+
+- Modern400 uses 400 identical unseen English prompts per system: 200 fixed modern/stress prompts plus 200 deterministic FLEURS `en_us` test prompts.
+- Exact-text exclusion was checked against 87,362 training transcripts.
+- Semantic WER is reported independently for `Qwen/Qwen3-ASR-1.7B`, `nvidia/nemotron-3.5-asr-streaming-0.6b`, and `openai/whisper-large-v3`.
+- All ASR inputs are explicitly resampled to 16 kHz and scored with the same disclosed English normalizer.
+- UTMOS22 uses `tarepan/SpeechMOS` v1.2.0 on a separate 500-prompt frozen generation set.
+- Headline intervals use 10,000 bootstrap samples.
+- The Modern400 corpus SHA-256 is `b7504ce2dce44a2da82770a6a5dfd2a034fe17e2113980f8a69663ade417a34c`.
+- Prompts, hypotheses, compressed row-level reports, and summaries ship under `evaluation/final/`.
+
+</details>
 
-- punctuation-preserving English normalization and stress-marked phonemes
-- compact Transformer text encoder and duration model
-- latent prior and residual coupling flow
-- integrated adversarial waveform decoder at 24 kHz
-- inference-only checkpoint with training discriminators and optimizer state removed
+## Runtime methodology
 
-| Component | Value |
+The model-specific snapshot below is retained as a transparent engineering measurement, not a portable hardware promise. It used one isolated process on an AMD EPYC 7K62, 50 prompts, three warmups, and runtime-default CPU threading.
+
+| RTF ↓ | Audio generated / wall time ↑ | Median utterance | p95 utterance | Cold load |
+| ---: | ---: | ---: | ---: | ---: |
+| 0.743 | 1.35× real time | 3.205 s | 4.681 s | 1.35 s |
+
+**RTF is wall time divided by generated audio duration.** For example, `0.54 RTF` means 1.85 seconds of audio were generated per second of wall time. It does not mean “1.85× faster than another model.” CPU model, thread policy, framework provider, and text length can materially change the result.
+
+<details>
+<summary><strong>Architecture and parameter budget</strong></summary>
+
+Inflect v2 is a parameter-efficient VITS-family, end-to-end text-to-waveform generator with an English phoneme frontend, monotonic alignment, stochastic latent synthesis, residual coupling flow, and an integrated alias-reduced neural waveform decoder.
+
+| Component | Inflect-Micro-v2 |
 | --- | ---: |
 | Latent channels | 192 |
 | Text hidden channels | 96 |
-| Text encoder layers / heads | 3 / 2 |
+| Encoder layers / heads | 3 / 2 |
 | Feed-forward channels | 768 |
 | Flow coupling blocks | 4 |
 | Initial decoder channels | 320 |
 | Upsample rates | 8, 8, 2, 2 |
 | Training segment | 16,384 samples |
+| Output | 24 kHz mono waveform |
 
-The model card describes the deployable architecture. The private corpus-construction and optimization recipe is not part of this open-weight release.
+The release describes the deployable architecture. Private corpus-construction and optimization details are not part of this open-weight package.
 
-## Runtime
+</details>
 
-Warm end-to-end CPU inference on **AMD EPYC 7K62 48-Core Processor**, with **192 logical CPUs visible** to the isolated process and runtime-default thread settings:
+<details>
+<summary><strong>Controls, determinism, and long text</strong></summary>
 
-| RTF ↓ | Audio / wall time ↑ | Median latency | p95 latency | Cold load |
-| ---: | ---: | ---: | ---: | ---: |
-| 0.7431 | 1.35× | 3.205 s | 4.681 s | 1.35 s |
+| Control | Default | Public range | Meaning |
+| --- | ---: | ---: | --- |
+| `speed` | `1.0` | `0.5–2.0` | Lower is slower; higher is faster. |
+| `variation` | `0.667` | `0.0–1.0` | Lower is steadier; higher samples more latent variation. |
+| `seed` | `0` | integer | Repeats the same stochastic sample on the same runtime stack. |
+
+Long passages are punctuation-aware chunks, not one unlimited autoregressive pass. Chunk boundaries receive short pauses and edge fades. See [`docs/API.md`](docs/API.md) for waveform contracts and concurrency notes.
+
+</details>
 
-Runtime varies with processor, PyTorch build, thread policy, text length, and operating system. Use the raw report rather than extrapolating these measurements to untested devices.
+<details>
+<summary><strong>Data, voice, and adaptation status</strong></summary>
+
+The release voice was trained on a single-speaker synthetic English corpus generated with a larger third-party TTS teacher. No teacher model is required or shipped at inference. The package does not redistribute a real speaker dataset and does not claim the synthetic voice as the identity of a real person.
 
-## Fine-tuning
+This release is inference-first. New-voice and new-language adaptation are **not currently validated or supported**. A new voice would replace the fixed speaker rather than add a selectable speaker; language adaptation also requires rebuilding normalization, phonemes, symbols, embeddings, and training data. See [`docs/DATA_AND_VOICE.md`](docs/DATA_AND_VOICE.md) and [`docs/FINETUNING.md`](docs/FINETUNING.md).
 
-The initial release is inference-first. The checkpoint is structurally compatible with voice and language adaptation, but the private training recipe and corpus-generation pipeline are not included. A tested public adaptation toolkit should be treated as a separate release rather than inferred from the inference package. See `docs/FINETUNING.md` for the supported status and data contract.
+</details>
 
-## Repository Layout
+## Package map
 
 | Path | Purpose |
 | --- | --- |
 | `model.pth` | Inference-only generator checkpoint |
-| `config.json` | Deployable architecture and audio configuration |
+| `config.json` | Architecture and audio configuration; also the Hub download-count query file |
 | `inference.py` | Public Python API and CLI |
-| `inflect_vits_frontend.py` | Normalization, phonemization, and punctuation frontend |
-| `runtime/` | Self-contained model implementation required for inference |
+| `inflect_vits_frontend.py` | English normalization, phonemization, and punctuation frontend |
+| `runtime/` | Self-contained model implementation |
 | `samples/` | Held-out example generations |
-| `evaluation/final/` | Final matched benchmark reports and protocol artifacts |
-| `release_manifest.json` | File sizes and SHA-256 integrity hashes |
+| `evaluation/final/` | Frozen benchmark prompts, reports, and protocol artifacts |
+| `docs/` | API, deployment, evaluation, data, and adaptation documentation |
+| `release_manifest.json` | File sizes and SHA-256 hashes |
 
 ## Limitations
 
-- English only and one fixed male voice. This is not a zero-shot voice-cloning model.
-- Unfamiliar phrasing can become flatter or less expressive.
-- Numbers, abbreviations, homographs, and uncommon names depend strongly on frontend normalization and context.
-- Long passages are synthesized in chunks, so transitions can differ from a single long-form model pass.
-- Stochastic variation can change pronunciation and timing; use a fixed seed for reproducibility.
-- Learned MOS predictors can reward or penalize spectral traits differently from human listeners.
+- English only, with one fixed male voice. This is not zero-shot voice cloning.
+- Unfamiliar phrasing can become flatter, less expressive, or less stable.
+- Numbers, abbreviations, homographs, and uncommon names remain frontend- and context-sensitive.
+- Long passages use punctuation-aware chunking; transitions can differ from a native long-form model pass.
+- Stochastic variation can alter timing and pronunciation. Fix the seed for comparisons.
+- UTMOS22 and ASR scores do not replace controlled human MOS or MUSHRA-style evaluation.
 - Not validated for medical, legal, emergency, or accessibility-critical communication.
 
-## Responsible Use
+## Responsible use
 
-Do not use the included voice to impersonate a real person, deceive listeners, or create fraudulent content. Disclose synthetic speech where the context could otherwise mislead. Users are responsible for applicable laws and the license.
+Do not use the included voice to impersonate a real person, deceive listeners, or create fraudulent content. Disclose synthetic speech where the context could otherwise mislead. Users are responsible for applicable laws and the Apache-2.0 license.
 
-## Integrity and License
+## License, integrity, and attribution
 
-`release_manifest.json` records file sizes and SHA-256 hashes. Original Inflect code and weights in this repository are licensed under Apache-2.0; bundled third-party components retain their own notices in `THIRD_PARTY_NOTICES.md`.
+Original Inflect code and weights are released under Apache-2.0. Bundled third-party components retain their own notices in [`THIRD_PARTY_NOTICES.md`](THIRD_PARTY_NOTICES.md). `release_manifest.json` records packaged file sizes and SHA-256 hashes.
 
 ## Citation
 
@@ -226,3 +277,5 @@ Do not use the included voice to impersonate a real person, deceive listeners, o
   url = {https://huggingface.co/owensong/Inflect-Micro-v2}
 }
 ```
+
+<p align="center"><sub>Developed independently by Owen Song. Inflect v2 is an open-weight release: inference code and deployable weights are public; the private training recipe and corpus-generation pipeline are not.</sub></p>

THIRD_PARTY_NOTICES.md

@@ -24,3 +24,13 @@ VITS. The text frontend also retains its original Keith Ito MIT license at
 
 The lightweight alias-free waveform activation implementation derives from
 BigVGAN's alias-free design and was adapted for this compact runtime.
+
+## alias-free-torch
+
+- Project: `alias-free-torch` (`junjun3518/alias-free-torch`)
+- License: Apache License 2.0
+- Packaged license: `third_party/ALIAS_FREE_TORCH_LICENSE.txt`
+
+The anti-aliased activation resampling design used by the compact waveform
+runtime includes concepts and adapted implementation structure from
+`alias-free-torch`.

docs/API.md

@@ -1,23 +1,98 @@
-# Python API
-
-## `InflectTTS(model_dir, device="cpu")`
-
-Loads one released checkpoint and its configuration. Construct the engine once and reuse it for multiple utterances.
-
-## `synthesize(text, *, speed=1.0, variation=0.667, seed=0)`
-
-Returns a mono `numpy.float32` waveform at 24 kHz. Empty text raises `ValueError`. `speed` must be between `0.5` and `2.0`; `variation` must be between `0.0` and `1.0`. Long input is split at punctuation-aware boundaries, and each chunk uses `seed + chunk_index`.
-
-## `save(text, output, **kwargs)`
-
-Synthesizes and writes a WAV file, creating parent directories as needed. It returns the output `Path`.
-
-```python
-from inference import InflectTTS
-
-tts = InflectTTS(".", device="cpu")
-waveform = tts.synthesize("A compact model can still speak clearly.", seed=7)
-tts.save("This is written directly to disk.", "out.wav", seed=7)
-```
-
-The API is intentionally small. It does not expose speaker cloning, streaming state, SSML, or language switching in the v2 launch package.
+# Python API
+
+## Load once
+
+```python
+from inference import InflectTTS
+
+tts = InflectTTS(model_dir=".", device="cpu")
+```
+
+`model_dir` must contain `model.pth`, `config.json`, the public frontend, and
+`runtime/`. `device` accepts a PyTorch device string such as `cpu`, `cuda`, or
+`cuda:0`. Loading validates the release architecture and raises
+`FileNotFoundError` for missing artifacts or `RuntimeError` for an incompatible
+checkpoint.
+
+Keep one engine alive for repeated requests. Constructing an engine reloads the
+checkpoint and should not happen per utterance.
+
+## `synthesize`
+
+```python
+sample_rate, waveform = tts.synthesize(
+    "A compact model can still speak clearly.",
+    speed=1.0,
+    variation=0.667,
+    seed=7,
+)
+```
+
+Signature:
+
+```python
+synthesize(text, *, speed=1.0, variation=0.667, seed=0)
+```
+
+Returns `(sample_rate, waveform)`:
+
+- `sample_rate`: integer `24000`, in samples per second.
+- `waveform`: one-dimensional, mono `numpy.ndarray` with `float32` samples
+  clipped to `[-1.0, 1.0]`.
+
+Validation:
+
+- empty or whitespace-only text raises `ValueError`;
+- `speed` must be in `0.5–2.0`;
+- `variation` must be in `0.0–1.0`;
+- `seed` is converted to an integer.
+
+`seed` repeats the same stochastic sample on the same model/runtime stack.
+Cross-version or cross-device bit identity is not guaranteed.
+
+## `save`
+
+```python
+output = tts.save(
+    "This is written directly to disk.",
+    "outputs/out.wav",
+    seed=7,
+)
+```
+
+`save(text, output, **synthesis_options)` creates parent directories, writes a
+mono 24 kHz PCM WAV, and returns the output `Path`.
+
+## Long text
+
+Long input is normalized, split first at sentence punctuation and then at safe
+intra-sentence boundaries. Each chunk is generated independently with
+`seed + chunk_index`, edge-faded, and joined with punctuation-dependent silence.
+This bounds memory, but it is not one globally planned long-form prosody pass.
+
+## Concurrency
+
+One `InflectTTS` instance mutates random-number state during synthesis. Serialize
+requests to one instance. For concurrent serving, use a worker pool with one
+engine per worker/device and enforce an application-level queue.
+
+## CLI
+
+```bash
+python inference.py \
+  --model-dir . \
+  --device cpu \
+  --text "Inflect is running locally." \
+  --output out.wav \
+  --speed 1.0 \
+  --variation 0.667 \
+  --seed 7
+```
+
+Run `python inference.py --help` for the authoritative flag list.
+
+## Deliberate omissions
+
+The v2 API does not expose speaker cloning, streaming state, SSML, native pitch
+control, or language switching. Pitch-shifting generated audio is
+post-processing and is not part of the model contract.

docs/DATA_AND_VOICE.md

@@ -0,0 +1,42 @@
+# Data and voice
+
+## Release scope
+
+Inflect v2 is a fixed-voice English text-to-speech release. It is not a
+zero-shot voice-cloning model and does not accept reference audio at inference.
+
+## Training-data summary
+
+The release model was trained on a single-speaker synthetic English speech
+corpus generated with a larger third-party text-to-speech teacher. The private
+corpus-construction and optimization recipe is not distributed. The inference
+package does not ship or call the teacher model.
+
+The corpus was filtered for transcript consistency, duration, clipping, and
+other audio-quality failures before model training. Frozen unseen-text prompts
+were excluded from training and are included with the final evaluation
+artifacts.
+
+## Voice status
+
+The packaged voice is a synthetic model voice. This repository does not
+redistribute a real-speaker recording corpus and does not claim that the output
+is the identity of a real person. Users must not present generated speech as a
+real person's recording.
+
+## Expected biases
+
+- English normalization and phonemization are optimized for general North
+  American English.
+- Proper names, uncommon locations, abbreviations, homographs, and code-switched
+  text may be mispronounced.
+- Synthetic teacher data can transfer its rhythm, prosody, spectral traits, and
+  pronunciation biases to the student.
+- The single-speaker design does not represent demographic or accent diversity.
+
+## What is not disclosed
+
+The private corpus prompts, teacher configuration, sampling curriculum,
+filtering thresholds, and optimization schedule are not part of this open-weight
+release. Their absence does not change the deployable parameter count: all
+inference-time components and weights are included.

docs/DEPLOYMENT.md

@@ -1,8 +1,41 @@
-# Deployment notes
-
-## Installation
-
-Use Python 3.10 or newer in a clean virtual environment, install `requirements.txt`, then run `inference.py` from the model repository. The package is tested with PyTorch CPU and CUDA execution; device-specific performance depends on the local PyTorch build.
+# Deployment
+
+## Verified release environment
+
+The final package was validated on Windows x86-64 with Python 3.11,
+PyTorch 2.10, CPU inference, and CUDA inference. The runtime is ordinary
+PyTorch and is expected to work on modern Linux x86-64, but that combination is
+not yet part of the signed release matrix. Treat Python 3.11 as the reference
+environment rather than interpreting flexible requirements as proof of every
+possible version.
+
+Install the CPU or CUDA PyTorch wheel for the target first, then install either:
+
+```bash
+python -m pip install -r requirements.txt
+```
+
+or the exact validation set:
+
+```bash
+python -m pip install -r requirements-tested.txt
+```
+
+The PyTorch build suffix remains platform-specific and is intentionally not
+forced by the tested requirements file.
+
+## Revision-pinned download
+
+```python
+from huggingface_hub import snapshot_download
+
+model_dir = snapshot_download(
+    "owensong/Inflect-Micro-v2",
+    revision="<release-commit-sha>",
+)
+```
+
+Replace the placeholder with the final immutable Hub revision after launch.
 
 ## Reproducibility
 
@@ -12,13 +45,25 @@ Keep the checkpoint, `config.json`, frontend files, and `runtime/` directory tog
 
 The runtime splits long text at sentence punctuation and then at safe intra-sentence boundaries. Chunks are synthesized independently and joined with punctuation-dependent silence. This prevents unbounded inference but means a long paragraph is not one globally planned prosody pass.
 
-## Production boundaries
+## Production boundaries
 
 - Instantiate `InflectTTS` once and reuse it; loading per request wastes time.
 - Serialize calls to one model instance unless the surrounding application explicitly manages device memory and random seeds.
 - Validate empty or untrusted input before synthesis.
 - Set a fixed seed for deterministic caching and tests.
-- Measure on the actual target device before claiming real-time operation.
+- Measure on the actual target device before claiming real-time operation.
+- Apply an explicit CPU thread policy in production rather than inheriting all
+  visible host cores.
+
+## Capacity planning
+
+- The FP32 checkpoint is 37.53 MB for Micro and 15.97 MB for Nano.
+- Peak RAM and VRAM depend on PyTorch, allocator state, input length, and device.
+  Measure the deployed process rather than estimating memory from weight size.
+- The reference runtime is non-streaming: the full chunk is returned after
+  synthesis.
+- One model instance should process one request at a time unless the application
+  owns synchronization and random seed isolation.
 
 ## Troubleshooting
 
@@ -26,4 +71,4 @@ The runtime splits long text at sentence punctuation and then at safe intra-sent
 - Unexpected pronunciation: spell out ambiguous abbreviations or add context; the release frontend is English-only.
 - Flat or unstable delivery: keep `variation` near the default before changing speed.
 - Long pauses: inspect source punctuation because chunk boundaries intentionally follow punctuation.
-- Slow first request: model loading is measured separately from warm synthesis and should not happen per utterance.
+- Slow first request: model loading is measured separately from warm synthesis and should not happen per utterance.

docs/EVALUATION.md

@@ -23,9 +23,20 @@ Semantic WER is run independently with Qwen3-ASR-1.7B, Nemotron 3.5 ASR 0.6B, an
 
 UTMOS22 is run over every generated clip with 10,000-sample bootstrap intervals. It is a learned quality predictor, not human MOS, and is reported separately from WER and human preference.
 
-## Runtime
-
-Runtime reports use one named host, the same 50-prompt subset, warm end-to-end synthesis, one isolated process per system, each public runtime's default thread/provider behavior, and separate model-load timing. RTF, audio-seconds per wall-second, median latency, p95 latency, and per-prompt rows are retained. Cross-machine numbers are not mixed.
+## Runtime
+
+The retained launch snapshot uses one named host, the same 50-prompt subset,
+warm synthesis, one isolated process per system, runtime-default thread
+behavior, and separate model-load timing. RTF, audio-seconds per wall-second,
+median latency, p95 latency, and per-prompt rows are retained.
+
+This snapshot is explicitly **host-specific engineering evidence**, not a
+controlled cross-framework CPU leaderboard. The host exposed 192 logical CPUs,
+thread counts were not pinned, and framework endpoints differed. The model card
+therefore explains Inflect's own RTF but does not use the old competitor runtime
+chart as headline evidence. A publication-grade rerun requires fixed affinity
+and thread counts, active-provider capture, equivalent waveform endpoints,
+multiple randomized trials, and confidence intervals.
 
 ## Human evidence
 

docs/FINETUNING.md

@@ -1,4 +1,4 @@
-# Fine-tuning and adaptation status
+# Adaptation status
 
 `Inflect-Micro-v2` is an open-weight, inference-first release. The repository does not ship the private corpus-construction recipe, teacher setup, discriminator checkpoints, optimizer state, or a supported training command.
 
@@ -7,14 +7,16 @@
 - Local inference from the released generator checkpoint.
 - Deterministic regression testing through a fixed text, seed, speed, and variation value.
 - Inspection of the complete deployable architecture and frontend.
-- Research adaptation in a separate VITS-compatible training environment, at the user's own risk.
+- Inspection of the released generator as a starting point for independent
+  research in a separately reconstructed VITS-compatible training stack.
 
 ## What is not yet supported
 
 - A one-command new-voice trainer.
 - Zero-shot voice cloning or speaker embeddings.
 - Language adaptation without rebuilding the symbol inventory and frontend.
-- Resuming the release checkpoint as if it contained discriminators or optimizer state.
+- Resuming the release checkpoint as if it contained discriminators or optimizer state.
+- Adding a selectable second speaker to this single-speaker architecture.
 
 ## Data contract for future adapters
 
@@ -24,4 +26,9 @@ A defensible voice-adaptation toolkit must require a UTF-8 manifest with one aud
 
 Any adapted checkpoint should be compared with the frozen release on identical unseen prompts, a fixed seed policy, semantic WER, signal diagnostics, a learned quality predictor, and a blinded human listening test. Improving speaker similarity while reducing intelligibility is not a successful adaptation.
 
-This file intentionally does not pretend that an unvalidated training recipe is a supported feature. A public adapter will be documented separately if it passes the same package and evaluation gates as the release models.
+The released checkpoint may be technically reusable inside a reconstructed
+training system, but new-voice and new-language adaptation have not been
+validated. This file intentionally does not turn theoretical compatibility into
+a supported feature. A public adapter will be documented separately only if it
+can reproduce a usable checkpoint from a clean environment and pass the same
+package and evaluation gates as the release models.

release_manifest.json

@@ -7,21 +7,6 @@
   "smoke_audio_seconds": 3.968,
   "standalone_cpu_smoke_audio_seconds": 3.616,
   "files": [
-    {
-      "path": "assets/evidence/asr-robustness.svg",
-      "bytes": 4638,
-      "sha256": "3b2bdb670e95fd0f1899600f5456676ba3a1ed826066d12cb83474a24119fe88"
-    },
-    {
-      "path": "assets/evidence/category-semantic-wer.svg",
-      "bytes": 6123,
-      "sha256": "1a2218e0bbe726167a5b0201736a82947edbd7c6803df3f5689e7a711026758e"
-    },
-    {
-      "path": "assets/evidence/cpu-throughput.svg",
-      "bytes": 3612,
-      "sha256": "9628985af7ec40d05db1e32e4c00a209ae8e44abab7e5f76fbd4ddc5bec7a6a9"
-    },
     {
       "path": "assets/evidence/evidence-summary.json",
       "bytes": 52605,
@@ -34,28 +19,18 @@
     },
     {
       "path": "assets/evidence/model-footprint.svg",
-      "bytes": 3051,
-      "sha256": "83624b63bc7af81aa5b166a3a79a65c288314aaf9127be9010571116bc40bcf7"
+      "bytes": 2667,
+      "sha256": "10118b49c79f4054588d6ed79faa4d67f35d4e8f57fba6217ff20a94d0cc83b3"
     },
     {
       "path": "assets/evidence/modern400-three-asr.svg",
       "bytes": 13700,
       "sha256": "1f519806caedda4490b85a3bdac1e943766196622603220193a18b0572518b22"
     },
-    {
-      "path": "assets/evidence/quality-vs-cpu-speed.svg",
-      "bytes": 4390,
-      "sha256": "0482dddf453c849ff3587f3a1d70f247a26011cfe570ecd179397067674b9837"
-    },
     {
       "path": "assets/evidence/quality-vs-footprint.svg",
-      "bytes": 5054,
-      "sha256": "84ea1c4ff315854bfcb11790cd0204901a2991e6a7277e8a4faafb859b022a12"
-    },
-    {
-      "path": "assets/evidence/semantic-wer.svg",
-      "bytes": 4693,
-      "sha256": "0188c4ec4d244a9b90448da90d90c20f82ae09d0739cc593b8bb5bca04609bc9"
+      "bytes": 4039,
+      "sha256": "fa5905b9709ec8644b5876bbf5ae2c5f9e24548532d65d1bdcaefaa0f22d8b0f"
     },
     {
       "path": "assets/inflect-v2-repository-hero.png",
@@ -79,28 +54,28 @@
     },
     {
       "path": "docs/API.md",
-      "bytes": 1031,
-      "sha256": "e1047734797798eac06f94c7f6c87dec424cf675fd0dc1dc2e396547c2405c93"
+      "bytes": 2632,
+      "sha256": "1103418ce10351c0f58722d8139e9fdf3837a68a82ce2b4d9adb6b870eb2a9c3"
+    },
+    {
+      "path": "docs/DATA_AND_VOICE.md",
+      "bytes": 1739,
+      "sha256": "8a504473b8377493256d435124851479ebee7f74432671ed0e0dff4e4509d8ba"
     },
     {
       "path": "docs/DEPLOYMENT.md",
-      "bytes": 1883,
-      "sha256": "028d23cfcd6b474c3fee7318ccaf990bf153ba2060f6048918de2b6f870b98c1"
+      "bytes": 3208,
+      "sha256": "c66a4633757402a6f065435b6ac1f3ce7a41f2c8061771522ba435a947c2e2af"
     },
     {
       "path": "docs/EVALUATION.md",
-      "bytes": 2826,
-      "sha256": "32678834f86d540000ec9509f5c252beac30f313a19ffa0bfd5e3aaefa476a6d"
+      "bytes": 3288,
+      "sha256": "44ec8605f4981557a97ef41fc1b9105c5a96145ffef2b59b646ac001bdffc1a5"
     },
     {
       "path": "docs/FINETUNING.md",
-      "bytes": 1923,
-      "sha256": "c9bc195dae24d4326d9f7738acc8727bb73d0efee7880a4b200ac1ba82689f65"
-    },
-    {
-      "path": "evaluation/final/evidence-summary.json",
-      "bytes": 52605,
-      "sha256": "51797abe973df6a68962dac3dbff729706acd333e08c90e382910b6b6a85018e"
+      "bytes": 2241,
+      "sha256": "82c122acde5b6f7bcb675f16ec5ed573809137daf7ab9f8c970132837fdf60ba"
     },
     {
       "path": "evaluation/final/modern400/combined_manifest_validation.json",
@@ -128,120 +103,130 @@
       "sha256": "0dd338b3b3bf66908681777a43e8e755087135eb1c006ddc99b8aecbcf004f8b"
     },
     {
-      "path": "evaluation/final/raw/asr_combined/report.json",
+      "path": "evaluation/final/runtime_snapshot/inflect-micro-v2_cpu.json",
+      "bytes": 11741,
+      "sha256": "9c5bf61b71ffefba9e35cf2d3b6ed445d15be8a95e6dc5629d0555bff20e3d8e"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/inflect-micro-v2_cuda.json",
+      "bytes": 11876,
+      "sha256": "bd42a1fbed748b99036eb500e4c11766cfe14db91269816fa91547b9eb1b4b19"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/inflect-nano-v2_cpu.json",
+      "bytes": 11676,
+      "sha256": "ad45de5b45a478da98cbbaafebedfc8568ccafbc7ff6fcdd35e4f65a6191778f"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/inflect-nano-v2_cuda.json",
+      "bytes": 11835,
+      "sha256": "fc3666aff4b29e2dcda75e58792a8de3887f312909dfe2a8b856191fbfb280cb"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/kitten-nano-bruno_cpu.json",
+      "bytes": 11915,
+      "sha256": "c8932d2f11b63c7f4e002fed2775ac70d3b0b2f323f811330a181cb854dc25ff"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/kitten-nano-hugo_cpu.json",
+      "bytes": 11914,
+      "sha256": "a65be2e9b3620ec436eff45a548008a0e8a22fab28f208316c295ff471c421a5"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/piper-danny-low_cpu.json",
+      "bytes": 11320,
+      "sha256": "fa21a8f76ee193390a1e0bb25ed409ae2771c19ec4e874e53517c567bc2015a0"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/piper-ryan-low_cpu.json",
+      "bytes": 11316,
+      "sha256": "09627a21f90d93b2c46ead64c2ecced59e5dce0727bb6c6d1c13baa653fed13b"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/supertonic3-james-3step_cpu.json",
+      "bytes": 12005,
+      "sha256": "60c2928dd5baab704fd8f3d56dabc59a257681fb2ee3718e0d28a9d1266d0163"
+    },
+    {
+      "path": "evaluation/final/runtime_snapshot/supertonic3-james-8step_cpu.json",
+      "bytes": 11971,
+      "sha256": "f62ce39eef361aa5394bcd6083332c4c43189804eae86fe2a91b58de58b07993"
+    },
+    {
+      "path": "evaluation/final/utmos22/report.json",
+      "bytes": 761778,
+      "sha256": "ecac4dc3847da14ec56d6822480036f45cea49399875a4efdaf968fd2a258fac"
+    },
+    {
+      "path": "evaluation/legacy/release500/asr_combined/report.json",
       "bytes": 8810759,
       "sha256": "5ebacbe622cb0dccf278fe401f8e739b75e2ebf3025eb159b1224c315c1a1571"
     },
     {
-      "path": "evaluation/final/raw/combined_manifest_validation.json",
+      "path": "evaluation/legacy/release500/combined_manifest_validation.json",
       "bytes": 762,
       "sha256": "6e01c0a4973c979dc0ac92caa01b4d3928f16f7be1dd4cd5c1c652a6613a3872"
     },
     {
-      "path": "evaluation/final/raw/prompt_report.json",
+      "path": "evaluation/legacy/release500/evidence-summary.json",
+      "bytes": 52605,
+      "sha256": "51797abe973df6a68962dac3dbff729706acd333e08c90e382910b6b6a85018e"
+    },
+    {
+      "path": "evaluation/legacy/release500/prompt_report.json",
       "bytes": 867,
       "sha256": "bec272d41da5c2d27ba1249f9321dbc9af588ea053efa2a94ca79f91ac265263"
     },
     {
-      "path": "evaluation/final/raw/prompts.jsonl",
+      "path": "evaluation/legacy/release500/prompts.jsonl",
       "bytes": 75177,
       "sha256": "467245f6c03221e63ca0c4ed6d7c938a69efa38d9a03dd4f7dc62644523cd342"
     },
     {
-      "path": "evaluation/final/raw/signal_diagnostics/per_clip.csv",
+      "path": "evaluation/legacy/release500/signal_diagnostics/per_clip.csv",
       "bytes": 1949804,
       "sha256": "b82b853b63af8e80a0d237c967075977c74f47ead2dfa3f8bb222a7f22403e83"
     },
     {
-      "path": "evaluation/final/raw/signal_diagnostics/report.json",
+      "path": "evaluation/legacy/release500/signal_diagnostics/report.json",
       "bytes": 456895,
       "sha256": "d29f875fdf413b259e7e504dd2f4c35ecf85176b2cf186f3f49f1e49ebb0f0c9"
     },
     {
-      "path": "evaluation/final/raw/utmos22/report.json",
+      "path": "evaluation/legacy/release500/utmos22/report.json",
       "bytes": 761778,
       "sha256": "ecac4dc3847da14ec56d6822480036f45cea49399875a4efdaf968fd2a258fac"
     },
     {
-      "path": "evaluation/final/raw/wav2vec2_large_lv60k_960h/hypotheses_wav2vec2-large-lv60k-960h.json",
+      "path": "evaluation/legacy/release500/wav2vec2_large_lv60k_960h/hypotheses_wav2vec2-large-lv60k-960h.json",
       "bytes": 234660,
       "sha256": "1610687af59ec452dbcb59031f874f50cb3cbf5f69d06cff44ad030e2ff42969"
     },
     {
-      "path": "evaluation/final/raw/wav2vec2_large_lv60k_960h/report.json",
+      "path": "evaluation/legacy/release500/wav2vec2_large_lv60k_960h/report.json",
       "bytes": 4036222,
       "sha256": "11d0e28482a1293c33e7f2cff5a70d4587b8214ae14709e85b6e064c9a0c312f"
     },
     {
-      "path": "evaluation/final/raw/wav2vec2_large_lv60k_960h/rows.csv",
+      "path": "evaluation/legacy/release500/wav2vec2_large_lv60k_960h/rows.csv",
       "bytes": 1032392,
       "sha256": "719d1477b78ecc67ec5b98a6d7b5bc60be68d16f1952ea144fc4821018802fac"
     },
     {
-      "path": "evaluation/final/raw/whisper_large_v3/hypotheses_whisper-large-v3.json",
+      "path": "evaluation/legacy/release500/whisper_large_v3/hypotheses_whisper-large-v3.json",
       "bytes": 500178,
       "sha256": "dc90651ab76f771a030f4fa28e105d42692b05fe35e2f14c68aa14aa4735c08a"
     },
     {
-      "path": "evaluation/final/raw/whisper_large_v3/report.json",
+      "path": "evaluation/legacy/release500/whisper_large_v3/report.json",
       "bytes": 4776290,
       "sha256": "14fb31cafef64a9f28222cfb2c53d708112312142de96fe09d2423cfd7362a0c"
     },
     {
-      "path": "evaluation/final/raw/whisper_large_v3/rows.csv",
+      "path": "evaluation/legacy/release500/whisper_large_v3/rows.csv",
       "bytes": 1261719,
       "sha256": "76d586a85f7f4c6651a70eb83d49ef90c31e7a1f6b4ba200d55296fc898c3de9"
     },
-    {
-      "path": "evaluation/final/runtime/inflect-micro-v2_cpu.json",
-      "bytes": 11741,
-      "sha256": "9c5bf61b71ffefba9e35cf2d3b6ed445d15be8a95e6dc5629d0555bff20e3d8e"
-    },
-    {
-      "path": "evaluation/final/runtime/inflect-micro-v2_cuda.json",
-      "bytes": 11876,
-      "sha256": "bd42a1fbed748b99036eb500e4c11766cfe14db91269816fa91547b9eb1b4b19"
-    },
-    {
-      "path": "evaluation/final/runtime/inflect-nano-v2_cpu.json",
-      "bytes": 11676,
-      "sha256": "ad45de5b45a478da98cbbaafebedfc8568ccafbc7ff6fcdd35e4f65a6191778f"
-    },
-    {
-      "path": "evaluation/final/runtime/inflect-nano-v2_cuda.json",
-      "bytes": 11835,
-      "sha256": "fc3666aff4b29e2dcda75e58792a8de3887f312909dfe2a8b856191fbfb280cb"
-    },
-    {
-      "path": "evaluation/final/runtime/kitten-nano-bruno_cpu.json",
-      "bytes": 11915,
-      "sha256": "c8932d2f11b63c7f4e002fed2775ac70d3b0b2f323f811330a181cb854dc25ff"
-    },
-    {
-      "path": "evaluation/final/runtime/kitten-nano-hugo_cpu.json",
-      "bytes": 11914,
-      "sha256": "a65be2e9b3620ec436eff45a548008a0e8a22fab28f208316c295ff471c421a5"
-    },
-    {
-      "path": "evaluation/final/runtime/piper-danny-low_cpu.json",
-      "bytes": 11320,
-      "sha256": "fa21a8f76ee193390a1e0bb25ed409ae2771c19ec4e874e53517c567bc2015a0"
-    },
-    {
-      "path": "evaluation/final/runtime/piper-ryan-low_cpu.json",
-      "bytes": 11316,
-      "sha256": "09627a21f90d93b2c46ead64c2ecced59e5dce0727bb6c6d1c13baa653fed13b"
-    },
-    {
-      "path": "evaluation/final/runtime/supertonic3-james-3step_cpu.json",
-      "bytes": 12005,
-      "sha256": "60c2928dd5baab704fd8f3d56dabc59a257681fb2ee3718e0d28a9d1266d0163"
-    },
-    {
-      "path": "evaluation/final/runtime/supertonic3-james-8step_cpu.json",
-      "bytes": 11971,
-      "sha256": "f62ce39eef361aa5394bcd6083332c4c43189804eae86fe2a91b58de58b07993"
-    },
     {
       "path": "examples/basic.py",
       "bytes": 352,
@@ -284,13 +269,18 @@
     },
     {
       "path": "README.md",
-      "bytes": 11325,
-      "sha256": "74f7b95e9ae230e15286aee2af6a7b64744b8acf9e839b61ecf9a4315cc9b04c"
+      "bytes": 15293,
+      "sha256": "40a99e317b0efa8c35335f3b6448e8163b985b96647364a43198d6fe928563ef"
+    },
+    {
+      "path": "requirements-tested.txt",
+      "bytes": 306,
+      "sha256": "0e2580f1bd5ae75d37405addec35bca789c01b19f7ef75a03a061eb35bc23c7d"
     },
     {
       "path": "requirements.txt",
-      "bytes": 128,
-      "sha256": "51778f4f55911af92747707c7d1e61191f7af5381efea9b71cf5a1abb2efcb03"
+      "bytes": 150,
+      "sha256": "03d9af55e5beceee330a9d906c1e3ad2907552bbe0a4d21b7f47e0c7bb95cccd"
     },
     {
       "path": "runtime/attentions.py",
@@ -377,6 +367,11 @@
       "bytes": 192556,
       "sha256": "2b59ad3d449c22b851d4a5f882f7a5e9d3413d0bc1e1037aa3374a3470c91114"
     },
+    {
+      "path": "third_party/ALIAS_FREE_TORCH_LICENSE.txt",
+      "bytes": 11357,
+      "sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4"
+    },
     {
       "path": "third_party/BIGVGAN_LICENSE.txt",
       "bytes": 1097,
@@ -389,8 +384,8 @@
     },
     {
       "path": "THIRD_PARTY_NOTICES.md",
-      "bytes": 892,
-      "sha256": "ef5343cd39713d313aefcfd09bb55fd6606ae2dacabeee13e0abdaa2e5b62589"
+      "bytes": 1233,
+      "sha256": "2ead63b7d9cddd6c1d12a96d88de0ade6fd3eabd1278825c0a3a2c1d3647ab4b"
     }
   ],
   "format": "inflect_v2_release_manifest_v1"

requirements-tested.txt

@@ -0,0 +1,11 @@
+# Exact package versions used for the final local release validation.
+# Install the CPU or CUDA PyTorch wheel appropriate for your platform first.
+torch==2.10.0
+numpy==2.3.5
+scipy==1.16.3
+soundfile==0.13.1
+phonemizer==3.3.0
+espeakng-loader==0.2.4
+num2words==0.5.14
+Unidecode==1.4.0
+huggingface-hub==0.36.2

requirements.txt

@@ -1,4 +1,5 @@
 torch>=2.6
+huggingface-hub>=0.36
 numpy>=1.26,<3
 scipy>=1.13
 soundfile>=0.13

third_party/ALIAS_FREE_TORCH_LICENSE.txt

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.