kale-tts-1-data / README.md
NERDDISCO's picture
kale-tts-1
3db59cb
|
Raw History Blame Contribute Delete
13.9 kB
---
license: cc-by-4.0
language: [en]
pretty_name: kale-tts-1-data
task_categories: [text-to-speech]
size_categories: [10K<n<100K]
annotations_creators: [machine-generated]
language_creators: [found] # texts: public-domain books
source_datasets: [original]
tags: [synthetic-speech, designed-voices, based-on-maya, tts]
configs:
- config_name: default
data_files:
- split: train
path: data/*/train-*.parquet
- split: test
path: data/*/test-*.parquet
- config_name: preview
data_files:
- split: test
path: preview/*.parquet
- {config_name: s057b, data_files: [{split: train, path: data/s057b/train-*.parquet}, {split: test, path: data/s057b/test-*.parquet}]}
- {config_name: f06, data_files: [{split: train, path: data/f06/train-*.parquet}, {split: test, path: data/f06/test-*.parquet}]}
- {config_name: d03, data_files: [{split: train, path: data/d03/train-*.parquet}, {split: test, path: data/d03/test-*.parquet}]}
- {config_name: w165, data_files: [{split: train, path: data/w165/train-*.parquet}, {split: test, path: data/w165/test-*.parquet}]}
- {config_name: w602, data_files: [{split: train, path: data/w602/train-*.parquet}, {split: test, path: data/w602/test-*.parquet}]}
- {config_name: m3, data_files: [{split: train, path: data/m3/train-*.parquet}, {split: test, path: data/m3/test-*.parquet}]}
- {config_name: t071, data_files: [{split: train, path: data/t071/train-*.parquet}, {split: test, path: data/t071/test-*.parquet}]}
- {config_name: d13, data_files: [{split: train, path: data/d13/train-*.parquet}, {split: test, path: data/d13/test-*.parquet}]}
- {config_name: x328, data_files: [{split: train, path: data/x328/train-*.parquet}, {split: test, path: data/x328/test-*.parquet}]}
- {config_name: h013f, data_files: [{split: train, path: data/h013f/train-*.parquet}, {split: test, path: data/h013f/test-*.parquet}]}
---
![kale-tts-1-data](diagrams/header-data.png)
# kale-tts-1-data
kale-tts-1-data is 204.7 hours of original [Maya-1](https://huggingface.co/maya-research/maya1) speech in 10 designed English voices,
each take with the text it reads. It's made for training small text-to-speech models: it trained [kale-tts-1-base](https://huggingface.co/KaleComputer/kale-tts-1-base) and our two models,
[kale-tts-1-nano](https://huggingface.co/KaleComputer/kale-tts-1-nano) and [kale-tts-1-tiny](https://huggingface.co/KaleComputer/kale-tts-1-tiny).
Every voice was designed from a one-line text description; nobody was recorded and nobody was cloned.
| measure | value |
|---|---|
| audio | 204.7 hours, 99,783 takes, 24 kHz mono FLAC |
| voices | 10 (5 women, 5 men), US and British English |
| splits | train 99,166 takes (203.5 h), test 617 takes (1.2 h), preview 200 takes |
| texts | 72,746 single sentences and 27,037 passages of 2-5 sentences, from 55 public-domain books |
| per take | audio, Maya-1's codes, text, voice prompt, book, pace, pitch, Whisper read-back, voice similarity |
| trained our models | 93,563 takes, 192.8 hours (column `in_release_training`) |
| size | 19.9 GB in 55 Parquet files |
| licence | CC BY 4.0 |
## Voices
| name | voice | pronunciation | config | takes | hours |
|---|---|---|---|---|---|
| Kenzie | esports hype woman | US English | `s057b` | 8,754 | 17.25 |
| Simone | deep and confident woman | US English | `f06` | 12,195 | 28.51 |
| Amy | crisp news-style woman | British English | `d03` | 7,622 | 17.97 |
| Frankie | hard-boiled detective's partner | US English | `w165` | 7,374 | 17.08 |
| Darlene | rough low-voiced old trucker woman | US English | `w602` | 7,239 | 16.90 |
| Owen | bright young host | US English | `m3` | 12,289 | 23.18 |
| Toby | mischievous prankster | US English | `t071` | 9,958 | 19.37 |
| Jack | crisp news-style man | British English | `d13` | 17,441 | 33.82 |
| Earl | African American jazz granddad | US English | `x328` | 7,997 | 15.02 |
| Magnus | deep dramatic trailer narrator | US English | `h013f` | 8,914 | 15.59 |
Each voice is one voice prompt given to Maya-1. Kenzie's: "Esports caster, female voice in her 20s with an American accent. Bright
sharp timbre, very fast pacing, explosive excited tone at high intensity." All ten are in `metadata/voices.json`, and every take carries
its own in the `description` column (Owen has two: one with "brisk pacing", one with "conversational pacing").
## Listen
The `preview` config has 20 test takes of each voice, 200 in all, small enough for the dataset viewer above. To hear the voices
speak any text, try [the kale-tts-1 demo](https://huggingface.co/spaces/KaleComputer/kale-tts-1).
```python
from datasets import load_dataset
preview = load_dataset("KaleComputer/kale-tts-1-data", "preview", split="test")
take = preview[0]
print(take["voice"], take["text"])
audio = take["audio"] # 24 kHz, decoded by datasets
```
## How we made it
![How kale-tts-1-data was made](diagrams/datamade.png)
1. **Voice prompt.** One line of text per voice (above). Maya-1 makes a voice from a description only; there is no reference audio
anywhere in the chain.
2. **Render.** Maya-1 ([maya-research/maya1](https://huggingface.co/maya-research/maya1) @ `b1b9b38c`) reads each sentence or passage
once in the voice, with the prompt `<description="..."> text`. It runs on vLLM 0.33.0 in bf16 on Modal, 4 H100 GPUs in parallel
(one per container, 8 CPU cores and 64 GiB each), 10,000 prompts per call. vLLM batches all of a call's prompts on the GPU at once:
about 14x faster than plain generation, 255x real time per H100, about $0.02 of GPU time per hour of audio. Loading the model takes
80-110 s, so big calls are what keep it cheap. 46,000 takes (118 hours of audio) took about 17 minutes on 4 H100s. Sampling:
temperature 0.4, top-p 0.9, repetition penalty 1.1, one seed per take. SNAC 24 kHz
([hubertsiuzdak/snac_24khz](https://huggingface.co/hubertsiuzdak/snac_24khz) @ `d73ad176`) turns Maya-1's codes into audio.
3. **Filter.** Every take is measured and kept only if it passes every check (next sections). The measuring ran on up to 8 GPUs
(H100 or L40S), the rough-ending measure on up to 64 CPU containers.
4. **Books.** Only takes from books that are clearly public domain in the EU and the US go in.
5. **Split.** Sentences that appear together in any take form one group; a group goes to `test` by a hash of its text, so no `test`
sentence is spoken anywhere in `train`, in any voice.
The audio is Maya-1's output as it came out: no denoising, trimming or loudness change. The scripts are in the `recipe/render/` and
`recipe/filter/` folders of [kale-tts-1-tiny](https://huggingface.co/KaleComputer/kale-tts-1-tiny).
## What's in it
![What's in kale-tts-1-data](diagrams/datacontents.png)
- `data/<voice>/train-NNNNN-of-MMMMM.parquet` and `data/<voice>/test-00000-of-00001.parquet`: 55 files, each up to about 0.5 GB, 19.9 GB in all.
- `preview/preview-00000-of-00001.parquet`: the 200 preview takes.
- `metadata/metadata.parquet` (12.6 MB): every column except `audio` and `maya_codes`, for filtering without downloading audio.
- `metadata/books.json`: the source books with the death years that decide their status. `metadata/voices.json`: the voice prompts.
| column | meaning |
|---|---|
| `audio` | 24 kHz mono FLAC (16-bit), converted from Maya-1 + SNAC's float32 output; every take peaks at or below 0.975, so nothing clips |
| `maya_codes` | the take's raw SNAC codes as Maya-1 produced them (frames x 7, uint16, token id minus 128266); SNAC 24 kHz decodes them back to the original float audio |
| `id`, `voice`, `split` | take id, voice id (the config name above), split |
| `text` | what was read |
| `kind`, `n_sentences` | `single` (one sentence) or `passage` (2-5 consecutive sentences of one paragraph), and how many sentences |
| `description` | the voice prompt the take was rendered from |
| `book_id`, `book_title`, `book_author` | the source book |
| `seconds`, `wpm`, `f0_median`, `voiced_share`, `peak` | length, spoken words a minute, median pitch in Hz, voiced share, peak level |
| `lead_silence_s`, `tail_silence_s`, `end_level_db` | silence before and after the speech, level of the last frames (a cut-off take ends loud) |
| `whisper_text`, `whisper_missed_words`, `whisper_extra_words` | Whisper large-v3-turbo's read-back of the take |
| `end_token` | Maya-1 ended the take with its own end token (filled in where the render logged it) |
| `voice_similarity` | ECAPA-TDNN (`speechbrain/spkrec-ecapa-voxceleb`) cosine of the take to its voice's centre, on the released audio; higher = closer |
| `render_round` | the render round the take came from |
| `in_release_training` | the take trained kale-tts-1-nano and kale-tts-1-tiny |
## How we selected the takes
A take is kept only if all of these hold:
| check | rule |
|---|---|
| complete | Maya-1's own end token |
| read-back | Whisper large-v3-turbo hears exactly the text: no missed and no extra words |
| edges | 20 ms to 0.8 s of silence before, at most 0.8 s after; not cut off (last frames at or below -15 dB); no full-scale sample |
| pace | 100-230 spoken words a minute (Kenzie: up to 270, her natural pace) |
| pitch | median pitch within 15 % of the voice's median |
| voice match | a speaker embedding close to the voice's own centre (the mean of its closest 60 % of takes) |
| sentence ends | no rough (creaky or cut-off) ending: under 100 ms of rough frames in the last ~300 ms of every sentence, inside passages too; sentence ends found by forced alignment; pitch floor 110 Hz for women, 70 Hz for men. One rough inner end drops the whole passage. |
Most takes don't make it. In the render rounds behind this data, the filter kept 115,246 of 284,122 rendered takes (41 %). A take can
fail more than one check; the most common:
| check failed | takes |
|---|---|
| read-back: missed words | 71,077 |
| read-back: extra words | 70,846 |
| pace | 64,676 |
| rough ending | 58,012 |
| pitch | 17,987 |
| end edge | 5,185 |
| voice match | 1,368 |
| start edge | 1,217 |
| no end token | 581 |
| cut off | 313 |
Men need more renders: their takes fail the sentence-end check more often. After the filter, the book rule leaves out 18,551 more kept
takes (40.59 h) from six translated books whose translators' public-domain status we can't show. Rejected takes are not in this dataset.
## Use it
```python
from datasets import load_dataset
ds = load_dataset("KaleComputer/kale-tts-1-data") # all 10 voices, train + test
jack = load_dataset("KaleComputer/kale-tts-1-data", "d13", split="train", streaming=True) # one voice, streamed
for take in jack.take(3):
print(take["text"], take["seconds"], take["wpm"])
```
Filter first without downloading audio:
```python
import pandas as pd
meta = pd.read_parquet("hf://datasets/KaleComputer/kale-tts-1-data/metadata/metadata.parquet")
fast = meta[(meta.voice == "s057b") & (meta.wpm <= 230)]
```
Train on `train`, evaluate on `test`: its sentences never appear in `train`. About 25 hours of clean takes per voice is where a small
model's clarity levels off; 5 hours already sounds right but is less clear. Train on whole passages and speak sentence by sentence:
passages carry the melody. `maya_codes` lets you train codec-token models on the exact Maya-1 tokens without re-encoding.
What it trained:
| model | parameters | CPU file (int8) | WebGPU file (float32) | Galaxy Tab S10 Lite | unclear words vs Maya-1 |
|---|---|---|---|---|---|
| [kale-tts-1-base](https://huggingface.co/KaleComputer/kale-tts-1-base) | 81.8M | 327 MB (PyTorch) | - | not built for the tablet | +3.9 points |
| [kale-tts-1-tiny](https://huggingface.co/KaleComputer/kale-tts-1-tiny) | 45.7M | 79.9 MB | 183.2 MB | 24x real time | +1.4 points |
| [kale-tts-1-nano](https://huggingface.co/KaleComputer/kale-tts-1-nano) | 7.5M | 14.8 MB | 30.6 MB | 44x real time | +1.6 points |
kale-tts-1-base is our teacher: Kokoro-82M fine-tuned on these 10 voices. tiny keeps its planner, nano is distilled from it, and both
learn their sound straight from this audio.
## Limits
- English only, US and British accents. Book narration, mostly older prose: no dialogue, no numbers, little modern vocabulary.
The voices read text; they don't converse.
- One voice prompt per voice: no speaker variety inside a voice.
- The filter is automatic. A small share of takes may still carry a mispronunciation Whisper heard as right, or a rough ending; the
deep male voices keep some creak at sentence ends.
- Kenzie has 17.25 hours here; the models trained on 5.41 of them. Her rows go up to 270 words a minute, her trained rows stay within 230.
- The models also trained on 17,282 takes from the six held-back books; those takes aren't in this dataset.
- The accent words in the voice prompts are Maya-1's own vocabulary. The voices are what Maya-1 made of them, not recordings of any
community, and may not sound like real speakers of those accents.
## Licence and credits
Audio, codes and metadata: CC BY 4.0. Texts: public domain in the EU and the US: the deciding person (author, or translator for a
translation) died in 1955 or earlier, and the book was published before 1931.
AI-generated speech: every take is synthetic speech made by an AI model, in voices that belong to no real person. Label anything made
from this data as AI-generated, never present it as a human recording, and never use it to imitate a real person.
Credits: based on Maya-1: all audio is original speech from Maya-1 (Maya Research, Apache-2.0), decoded with SNAC (Hubert Siuzdak, MIT).
Used for scoring only, nothing of them is in the data: Whisper large-v3-turbo (OpenAI, MIT) for the read-back, ECAPA-TDNN
`speechbrain/spkrec-ecapa-voxceleb` (SpeechBrain, Apache-2.0) for `voice_similarity`. Acknowledgements: the source books were made
available by Project Gutenberg (www.gutenberg.org).