Spaces:
Paused
Paused
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,117 +1,127 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
-
|
| 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 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
---
|
| 3 |
+
title: vansarah TTS
|
| 4 |
+
emoji: 🎙️
|
| 5 |
+
colorFrom: pink
|
| 6 |
+
colorTo: purple
|
| 7 |
+
sdk: gradio # Hoặc "streamlit" nếu dùng Streamlit
|
| 8 |
+
sdk_version: "4.28.3" # Phiên bản Gradio mới nhất
|
| 9 |
+
app_file: app.py
|
| 10 |
+
pinned: false
|
| 11 |
+
license: apache-2.0
|
| 12 |
+
language:
|
| 13 |
+
- en
|
| 14 |
+
base_model:
|
| 15 |
+
- yl4579/StyleTTS2-LJSpeech
|
| 16 |
+
pipeline_tag: text-to-speech
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
**vansarah** is an open-weight TTS model...
|
| 20 |
+
---
|
| 21 |
+
**vansarah** is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers comparable quality to larger models while being significantly faster and more cost-efficient. With Apache-licensed weights, vansarah can be deployed anywhere from production environments to personal projects.
|
| 22 |
+
|
| 23 |
+
<audio controls><source src="https://huggingface.co/mr-don88/vansarah-82M/resolve/main/samples/HEARME.wav" type="audio/wav"></audio>
|
| 24 |
+
|
| 25 |
+
🐈 **GitHub**: https://github.com/mr-don88/vansarah
|
| 26 |
+
|
| 27 |
+
🚀 **Demo**: https://huggingface.co/spaces/MR-DON/vansarah-tts
|
| 28 |
+
|
| 29 |
+
> [!NOTE]
|
| 30 |
+
> As of April 2025, the market rate of vansarah served over API is **under $1 per million characters of text input**, or under $0.06 per hour of audio output. (On average, 1000 characters of input is about 1 minute of output.) Sources: [ArtificialAnalysis/Replicate at 65 cents per M chars](https://artificialanalysis.ai/text-to-speech/model-family/vansarah#price) and [DeepInfra at 80 cents per M chars](https://deepinfra.com/mr-don88/vansarah-82M).
|
| 31 |
+
>
|
| 32 |
+
> This is an Apache-licensed model, and vansarah has been deployed in numerous projects and commercial APIs. We welcome the deployment of the model in real use cases.
|
| 33 |
+
|
| 34 |
+
> [!CAUTION]
|
| 35 |
+
> Fake websites like vansarahttsai_com (snapshot: https://archive.ph/nRRnk) and vansarahtts_net (snapshot: https://archive.ph/60opa) are likely scams masquerading under the banner of a popular model.
|
| 36 |
+
>
|
| 37 |
+
> Any website containing "vansarah" in its root domain (e.g. vansarahttsai_com, vansarahtts_net) is **NOT owned by and NOT affiliated with this model page or its author**, and attempts to imply otherwise are red flags.
|
| 38 |
+
|
| 39 |
+
- [Releases](#releases)
|
| 40 |
+
- [Usage](#usage)
|
| 41 |
+
- [EVAL.md](https://huggingface.co/mr-don88/vansarah-82M/blob/main/EVAL.md) ↗️
|
| 42 |
+
- [SAMPLES.md](https://huggingface.co/mr-don88/vansarah-82M/blob/main/SAMPLES.md) ↗️
|
| 43 |
+
- [VOICES.md](https://huggingface.co/mr-don88/vansarah-82M/blob/main/VOICES.md) ↗️
|
| 44 |
+
- [Model Facts](#model-facts)
|
| 45 |
+
- [Training Details](#training-details)
|
| 46 |
+
- [Creative Commons Attribution](#creative-commons-attribution)
|
| 47 |
+
- [Acknowledgements](#acknowledgements)
|
| 48 |
+
|
| 49 |
+
### Releases
|
| 50 |
+
|
| 51 |
+
| Model | Published | Training Data | Langs & Voices | SHA256 |
|
| 52 |
+
| ----- | --------- | ------------- | -------------- | ------ |
|
| 53 |
+
| **v1.0** | **2025 Jan 27** | **Few hundred hrs** | [**8 & 54**](https://huggingface.co/mr-don88/vansarah-82M/blob/main/VOICES.md) | `496dba11` |
|
| 54 |
+
| [v0.19](https://huggingface.co/mr-don88/kLegacy/tree/main/v0.19) | 2024 Dec 25 | <100 hrs | 1 & 10 | `3b0c392f` |
|
| 55 |
+
|
| 56 |
+
| Training Costs | v0.19 | v1.0 | **Total** |
|
| 57 |
+
| -------------- | ----- | ---- | ----- |
|
| 58 |
+
| in A100 80GB GPU hours | 500 | 500 | **1000** |
|
| 59 |
+
| average hourly rate | $0.80/h | $1.20/h | **$1/h** |
|
| 60 |
+
| in USD | $400 | $600 | **$1000** |
|
| 61 |
+
|
| 62 |
+
### Usage
|
| 63 |
+
You can run this basic cell on [Google Colab](https://colab.research.google.com/). [Listen to samples](https://huggingface.co/mr-don88/vansarah-82M/blob/main/SAMPLES.md). For more languages and details, see [Advanced Usage](https://github.com/mr-don88/vansarah?tab=readme-ov-file#advanced-usage).
|
| 64 |
+
```py
|
| 65 |
+
!pip install -q vansarah>=0.9.2 soundfile
|
| 66 |
+
!apt-get -qq -y install espeak-ng > /dev/null 2>&1
|
| 67 |
+
from vansarah import KPipeline
|
| 68 |
+
from IPython.display import display, Audio
|
| 69 |
+
import soundfile as sf
|
| 70 |
+
import torch
|
| 71 |
+
pipeline = KPipeline(lang_code='a')
|
| 72 |
+
text = '''
|
| 73 |
+
[vansarah](/kˈOkəɹO/) is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers comparable quality to larger models while being significantly faster and more cost-efficient. With Apache-licensed weights, [vansarah](/kˈOkəɹO/) can be deployed anywhere from production environments to personal projects.
|
| 74 |
+
'''
|
| 75 |
+
generator = pipeline(text, voice='af_heart')
|
| 76 |
+
for i, (gs, ps, audio) in enumerate(generator):
|
| 77 |
+
print(i, gs, ps)
|
| 78 |
+
display(Audio(data=audio, rate=24000, autoplay=i==0))
|
| 79 |
+
sf.write(f'{i}.wav', audio, 24000)
|
| 80 |
+
```
|
| 81 |
+
Under the hood, `vansarah` uses [`misaki`](https://pypi.org/project/misaki/), a G2P library at https://github.com/mr-don88/misaki
|
| 82 |
+
|
| 83 |
+
### Model Facts
|
| 84 |
+
|
| 85 |
+
**Architecture:**
|
| 86 |
+
- StyleTTS 2: https://arxiv.org/abs/2306.07691
|
| 87 |
+
- ISTFTNet: https://arxiv.org/abs/2203.02395
|
| 88 |
+
- Decoder only: no diffusion, no encoder release
|
| 89 |
+
|
| 90 |
+
**Architected by:** Li et al @ https://github.com/yl4579/StyleTTS2
|
| 91 |
+
|
| 92 |
+
**Trained by**: `@rzvzn` on Discord
|
| 93 |
+
|
| 94 |
+
**Languages:** Multiple
|
| 95 |
+
|
| 96 |
+
**Model SHA256 Hash:** `496dba118d1a58f5f3db2efc88dbdc216e0483fc89fe6e47ee1f2c53f18ad1e4`
|
| 97 |
+
|
| 98 |
+
### Training Details
|
| 99 |
+
|
| 100 |
+
**Data:** vansarah was trained exclusively on **permissive/non-copyrighted audio data** and IPA phoneme labels. Examples of permissive/non-copyrighted audio include:
|
| 101 |
+
- Public domain audio
|
| 102 |
+
- Audio licensed under Apache, MIT, etc
|
| 103 |
+
- Synthetic audio<sup>[1]</sup> generated by closed<sup>[2]</sup> TTS models from large providers<br/>
|
| 104 |
+
[1] https://copyright.gov/ai/ai_policy_guidance.pdf<br/>
|
| 105 |
+
[2] No synthetic audio from open TTS models or "custom voice clones"
|
| 106 |
+
|
| 107 |
+
**Total Dataset Size:** A few hundred hours of audio
|
| 108 |
+
|
| 109 |
+
**Total Training Cost:** About $1000 for 1000 hours of A100 80GB vRAM
|
| 110 |
+
|
| 111 |
+
### Creative Commons Attribution
|
| 112 |
+
|
| 113 |
+
The following CC BY audio was part of the dataset used to train vansarah v1.0.
|
| 114 |
+
|
| 115 |
+
| Audio Data | Duration Used | License | Added to Training Set After |
|
| 116 |
+
| ---------- | ------------- | ------- | --------------------------- |
|
| 117 |
+
| [Koniwa](https://github.com/koniwa/koniwa) `tnc` | <1h | [CC BY 3.0](https://creativecommons.org/licenses/by/3.0/deed.ja) | v0.19 / 22 Nov 2024 |
|
| 118 |
+
| [SIWIS](https://datashare.ed.ac.uk/handle/10283/2353) | <11h | [CC BY 4.0](https://datashare.ed.ac.uk/bitstream/handle/10283/2353/license_text) | v0.19 / 22 Nov 2024 |
|
| 119 |
+
|
| 120 |
+
### Acknowledgements
|
| 121 |
+
|
| 122 |
+
- 🛠️ [@yl4579](https://huggingface.co/yl4579) for architecting StyleTTS 2.
|
| 123 |
+
- 🏆 [@Pendrokar](https://huggingface.co/Pendrokar) for adding vansarah as a contender in the TTS Spaces Arena.
|
| 124 |
+
- 📊 Thank you to everyone who contributed synthetic training data.
|
| 125 |
+
- ❤️ Special thanks to all compute sponsors.
|
| 126 |
+
- 👾 Discord server: https://discord.gg/QuGxSWBfQy
|
| 127 |
+
- 🪽 vansarah is a Japanese word that translates to "heart" or "spirit". It is also the name of an [AI in the Terminator franchise](https://terminator.fandom.com/wiki/vansarah).
|