Spaces:
Sleeping
Sleeping
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,161 +1,171 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
language:
|
| 4 |
-
- en
|
| 5 |
-
base_model:
|
| 6 |
-
- yl4579/StyleTTS2-LJSpeech
|
| 7 |
-
pipeline_tag: text-to-speech
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
Kokoro
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
The
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
- [
|
| 153 |
-
- [
|
| 154 |
-
|
| 155 |
-
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 161 |
https://terminator.fandom.com/wiki/Kokoro
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
base_model:
|
| 6 |
+
- yl4579/StyleTTS2-LJSpeech
|
| 7 |
+
pipeline_tag: text-to-speech
|
| 8 |
+
title: Kokoro-80M Interface
|
| 9 |
+
sdk: gradio
|
| 10 |
+
emoji: 🐨
|
| 11 |
+
colorFrom: blue
|
| 12 |
+
colorTo: yellow
|
| 13 |
+
pinned: true
|
| 14 |
+
short_description: A interface made with gradio for Kokoro-82M model
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
<h1>This Gradio Interface was created by <a href="https://github.com/SamratBarai/">@SamratBarai</a></h1>
|
| 18 |
+
|
| 19 |
+
📣 Jan 12 Status: Intent to improve the base model https://hf.co/hexgrad/Kokoro-82M/discussions/36
|
| 20 |
+
|
| 21 |
+
❤️ Kokoro Discord Server: https://discord.gg/QuGxSWBfQy
|
| 22 |
+
|
| 23 |
+
<audio controls><source src="https://huggingface.co/hexgrad/Kokoro-82M/resolve/main/demo/HEARME.wav" type="audio/wav"></audio>
|
| 24 |
+
|
| 25 |
+
**Kokoro** is a frontier TTS model for its size of **82 million parameters** (text in/audio out).
|
| 26 |
+
|
| 27 |
+
On 25 Dec 2024, Kokoro v0.19 weights were permissively released in full fp32 precision under an Apache 2.0 license. As of 2 Jan 2025, 10 unique Voicepacks have been released, and a `.onnx` version of v0.19 is available.
|
| 28 |
+
|
| 29 |
+
In the weeks leading up to its release, Kokoro v0.19 was the #1🥇 ranked model in [TTS Spaces Arena](https://huggingface.co/hexgrad/Kokoro-82M#evaluation). Kokoro had achieved higher Elo in this single-voice Arena setting over other models, using fewer parameters and less data:
|
| 30 |
+
1. **Kokoro v0.19: 82M params, Apache, trained on <100 hours of audio**
|
| 31 |
+
2. XTTS v2: 467M, CPML, >10k hours
|
| 32 |
+
3. Edge TTS: Microsoft, proprietary
|
| 33 |
+
4. MetaVoice: 1.2B, Apache, 100k hours
|
| 34 |
+
5. Parler Mini: 880M, Apache, 45k hours
|
| 35 |
+
6. Fish Speech: ~500M, CC-BY-NC-SA, 1M hours
|
| 36 |
+
|
| 37 |
+
Kokoro's ability to top this Elo ladder suggests that the scaling law (Elo vs compute/data/params) for traditional TTS models might have a steeper slope than previously expected.
|
| 38 |
+
|
| 39 |
+
You can find a hosted demo at [hf.co/spaces/hexgrad/Kokoro-TTS](https://huggingface.co/spaces/hexgrad/Kokoro-TTS).
|
| 40 |
+
|
| 41 |
+
### Usage
|
| 42 |
+
|
| 43 |
+
The following can be run in a single cell on [Google Colab](https://colab.research.google.com/).
|
| 44 |
+
```py
|
| 45 |
+
# 1️⃣ Install dependencies silently
|
| 46 |
+
!git lfs install
|
| 47 |
+
!git clone https://huggingface.co/hexgrad/Kokoro-82M
|
| 48 |
+
%cd Kokoro-82M
|
| 49 |
+
!apt-get -qq -y install espeak-ng > /dev/null 2>&1
|
| 50 |
+
!pip install -q phonemizer torch transformers scipy munch
|
| 51 |
+
|
| 52 |
+
# 2️⃣ Build the model and load the default voicepack
|
| 53 |
+
from models import build_model
|
| 54 |
+
import torch
|
| 55 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 56 |
+
MODEL = build_model('kokoro-v0_19.pth', device)
|
| 57 |
+
VOICE_NAME = [
|
| 58 |
+
'af', # Default voice is a 50-50 mix of Bella & Sarah
|
| 59 |
+
'af_bella', 'af_sarah', 'am_adam', 'am_michael',
|
| 60 |
+
'bf_emma', 'bf_isabella', 'bm_george', 'bm_lewis',
|
| 61 |
+
'af_nicole', 'af_sky',
|
| 62 |
+
][0]
|
| 63 |
+
VOICEPACK = torch.load(f'voices/{VOICE_NAME}.pt', weights_only=True).to(device)
|
| 64 |
+
print(f'Loaded voice: {VOICE_NAME}')
|
| 65 |
+
|
| 66 |
+
# 3️⃣ Call generate, which returns 24khz audio and the phonemes used
|
| 67 |
+
from kokoro import generate
|
| 68 |
+
text = "How could I know? It's an unanswerable question. Like asking an unborn child if they'll lead a good life. They haven't even been born."
|
| 69 |
+
audio, out_ps = generate(MODEL, text, VOICEPACK, lang=VOICE_NAME[0])
|
| 70 |
+
# Language is determined by the first letter of the VOICE_NAME:
|
| 71 |
+
# 🇺🇸 'a' => American English => en-us
|
| 72 |
+
# 🇬🇧 'b' => British English => en-gb
|
| 73 |
+
|
| 74 |
+
# 4️⃣ Display the 24khz audio and print the output phonemes
|
| 75 |
+
from IPython.display import display, Audio
|
| 76 |
+
display(Audio(data=audio, rate=24000, autoplay=True))
|
| 77 |
+
print(out_ps)
|
| 78 |
+
```
|
| 79 |
+
If you have trouble with `espeak-ng`, see this [github issue](https://github.com/bootphon/phonemizer/issues/44#issuecomment-1540885186). [Mac users also see this](https://huggingface.co/hexgrad/Kokoro-82M/discussions/12#677435d3d8ace1de46071489), and [Windows users see this](https://huggingface.co/hexgrad/Kokoro-82M/discussions/12#67742594fdeebf74f001ecfc).
|
| 80 |
+
|
| 81 |
+
For ONNX usage, see [#14](https://huggingface.co/hexgrad/Kokoro-82M/discussions/14).
|
| 82 |
+
|
| 83 |
+
### Model Facts
|
| 84 |
+
|
| 85 |
+
No affiliation can be assumed between parties on different lines.
|
| 86 |
+
|
| 87 |
+
**Architecture:**
|
| 88 |
+
- StyleTTS 2: https://arxiv.org/abs/2306.07691
|
| 89 |
+
- ISTFTNet: https://arxiv.org/abs/2203.02395
|
| 90 |
+
- Decoder only: no diffusion, no encoder release
|
| 91 |
+
|
| 92 |
+
**Architected by:** Li et al @ https://github.com/yl4579/StyleTTS2
|
| 93 |
+
|
| 94 |
+
**Trained by**: `@rzvzn` on Discord
|
| 95 |
+
|
| 96 |
+
**Supported Languages:** American English, British English
|
| 97 |
+
|
| 98 |
+
**Model SHA256 Hash:** `3b0c392f87508da38fad3a2f9d94c359f1b657ebd2ef79f9d56d69503e470b0a`
|
| 99 |
+
|
| 100 |
+
### Releases
|
| 101 |
+
- 25 Dec 2024: Model v0.19, `af_bella`, `af_sarah`
|
| 102 |
+
- 26 Dec 2024: `am_adam`, `am_michael`
|
| 103 |
+
- 28 Dec 2024: `bf_emma`, `bf_isabella`, `bm_george`, `bm_lewis`
|
| 104 |
+
- 30 Dec 2024: `af_nicole`
|
| 105 |
+
- 31 Dec 2024: `af_sky`
|
| 106 |
+
- 2 Jan 2025: ONNX v0.19 `ebef4245`
|
| 107 |
+
|
| 108 |
+
### Licenses
|
| 109 |
+
- Apache 2.0 weights in this repository
|
| 110 |
+
- MIT inference code in [spaces/hexgrad/Kokoro-TTS](https://huggingface.co/spaces/hexgrad/Kokoro-TTS) adapted from [yl4579/StyleTTS2](https://github.com/yl4579/StyleTTS2)
|
| 111 |
+
- GPLv3 dependency in [espeak-ng](https://github.com/espeak-ng/espeak-ng)
|
| 112 |
+
|
| 113 |
+
The inference code was originally MIT licensed by the paper author. Note that this card applies only to this model, Kokoro. Original models published by the paper author can be found at [hf.co/yl4579](https://huggingface.co/yl4579).
|
| 114 |
+
|
| 115 |
+
### Evaluation
|
| 116 |
+
|
| 117 |
+
**Metric:** Elo rating
|
| 118 |
+
|
| 119 |
+
**Leaderboard:** [hf.co/spaces/Pendrokar/TTS-Spaces-Arena](https://huggingface.co/spaces/Pendrokar/TTS-Spaces-Arena)
|
| 120 |
+
|
| 121 |
+
|
| 122 |
+
|
| 123 |
+
The voice ranked in the Arena is a 50-50 mix of Bella and Sarah. For your convenience, this mix is included in this repository as `af.pt`, but you can trivially reproduce it like this:
|
| 124 |
+
|
| 125 |
+
```py
|
| 126 |
+
import torch
|
| 127 |
+
bella = torch.load('voices/af_bella.pt', weights_only=True)
|
| 128 |
+
sarah = torch.load('voices/af_sarah.pt', weights_only=True)
|
| 129 |
+
af = torch.mean(torch.stack([bella, sarah]), dim=0)
|
| 130 |
+
assert torch.equal(af, torch.load('voices/af.pt', weights_only=True))
|
| 131 |
+
```
|
| 132 |
+
|
| 133 |
+
### Training Details
|
| 134 |
+
|
| 135 |
+
**Compute:** Kokoro v0.19 was trained on A100 80GB vRAM instances for approximately 500 total GPU hours. The average cost for each GPU hour was around $0.80, so the total cost was around $400.
|
| 136 |
+
|
| 137 |
+
**Data:** Kokoro was trained exclusively on **permissive/non-copyrighted audio data** and IPA phoneme labels. Examples of permissive/non-copyrighted audio include:
|
| 138 |
+
- Public domain audio
|
| 139 |
+
- Audio licensed under Apache, MIT, etc
|
| 140 |
+
- Synthetic audio<sup>[1]</sup> generated by closed<sup>[2]</sup> TTS models from large providers<br/>
|
| 141 |
+
[1] https://copyright.gov/ai/ai_policy_guidance.pdf<br/>
|
| 142 |
+
[2] No synthetic audio from open TTS models or "custom voice clones"
|
| 143 |
+
|
| 144 |
+
**Epochs:** Less than **20 epochs**
|
| 145 |
+
|
| 146 |
+
**Total Dataset Size:** Less than **100 hours** of audio
|
| 147 |
+
|
| 148 |
+
### Limitations
|
| 149 |
+
|
| 150 |
+
Kokoro v0.19 is limited in some specific ways, due to its training set and/or architecture:
|
| 151 |
+
- [Data] Lacks voice cloning capability, likely due to small <100h training set
|
| 152 |
+
- [Arch] Relies on external g2p (espeak-ng), which introduces a class of g2p failure modes
|
| 153 |
+
- [Data] Training dataset is mostly long-form reading and narration, not conversation
|
| 154 |
+
- [Arch] At 82M params, Kokoro almost certainly falls to a well-trained 1B+ param diffusion transformer, or a many-billion-param MLLM like GPT-4o / Gemini 2.0 Flash
|
| 155 |
+
- [Data] Multilingual capability is architecturally feasible, but training data is mostly English
|
| 156 |
+
|
| 157 |
+
Refer to the [Philosophy discussion](https://huggingface.co/hexgrad/Kokoro-82M/discussions/5) to better understand these limitations.
|
| 158 |
+
|
| 159 |
+
**Will the other voicepacks be released?** There is currently no release date scheduled for the other voicepacks, but in the meantime you can try them in the hosted demo at [hf.co/spaces/hexgrad/Kokoro-TTS](https://huggingface.co/spaces/hexgrad/Kokoro-TTS).
|
| 160 |
+
|
| 161 |
+
### Acknowledgements
|
| 162 |
+
- [@yl4579](https://huggingface.co/yl4579) for architecting StyleTTS 2
|
| 163 |
+
- [@Pendrokar](https://huggingface.co/Pendrokar) for adding Kokoro as a contender in the TTS Spaces Arena
|
| 164 |
+
|
| 165 |
+
### Model Card Contact
|
| 166 |
+
|
| 167 |
+
`@rzvzn` on Discord. Server invite: https://discord.gg/QuGxSWBfQy
|
| 168 |
+
|
| 169 |
+
<img src="https://static0.gamerantimages.com/wordpress/wp-content/uploads/2024/08/terminator-zero-41-1.jpg" width="400" alt="kokoro" />
|
| 170 |
+
|
| 171 |
https://terminator.fandom.com/wiki/Kokoro
|