README.md
5.6 KB · 142 lines · markdown Raw
1 ---
2 license: mit
3 language:
4 - ar
5 - da
6 - de
7 - el
8 - en
9 - es
10 - fi
11 - fr
12 - he
13 - hi
14 - it
15 - ja
16 - ko
17 - ms
18 - nl
19 - no
20 - pl
21 - pt
22 - ru
23 - sv
24 - sw
25 - tr
26 - zh
27 pipeline_tag: text-to-speech
28 tags:
29 - text-to-speech
30 - speech
31 - speech-generation
32 - voice-cloning
33 - multilingual-tts
34 library_name: chatterbox
35 ---
36
37 <img width="800" alt="cb-big2" src="https://github.com/user-attachments/assets/bd8c5f03-e91d-4ee5-b680-57355da204d1" />
38
39 <h1 style="font-size: 32px">Chatterbox TTS</h1>
40
41 <div style="display: flex; align-items: center; gap: 12px">
42 <a href="https://resemble-ai.github.io/chatterbox_demopage/">
43 <img src="https://img.shields.io/badge/listen-demo_samples-blue" alt="Listen to Demo Samples" />
44 </a>
45 <a href="https://huggingface.co/spaces/ResembleAI/Chatterbox">
46 <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/open-in-hf-spaces-sm.svg" alt="Open in HF Spaces" />
47 </a>
48 <a href="https://podonos.com/resembleai/chatterbox">
49 <img src="https://static-public.podonos.com/badges/insight-on-pdns-sm-dark.svg" alt="Insight on Podos" />
50 </a>
51 </div>
52
53 <div style="display: flex; align-items: center; gap: 8px;">
54 <span style="font-style: italic;white-space: pre-wrap">Made with ❤️ by</span>
55 <img width="100" alt="resemble-logo-horizontal" src="https://github.com/user-attachments/assets/35cf756b-3506-4943-9c72-c05ddfa4e525" />
56 </div>
57
58 **09/04 🔥 Introducing Chatterbox Multilingual in 23 Languages!**
59
60 We're excited to introduce **Chatterbox** and **Chatterbox Multilingual**, [Resemble AI's](https://resemble.ai) production-grade open source TTS models. Chatterbox Multilingual supports **Arabic, Danish, German, Greek, English, Spanish, Finnish, French, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Dutch, Norwegian, Polish, Portuguese, Russian, Swedish, Swahili, Turkish, Chinese** out of the box. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.
61
62 Whether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support **emotion exaggeration control**, a powerful feature that makes your voices stand out. Try it now on our [Hugging Face Gradio app.](https://huggingface.co/spaces/ResembleAI/Chatterbox)
63
64 If you like the model but need to scale or tune it for higher accuracy, check out our competitively priced TTS service (<a href="https://resemble.ai">link</a>). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.
65
66 # Key Details
67 - Multilingual, zero-shot TTS supporting 23 languages
68 - SoTA zeroshot English TTS
69 - 0.5B Llama backbone
70 - Unique exaggeration/intensity control
71 - Ultra-stable with alignment-informed inference
72 - Trained on 0.5M hours of cleaned data
73 - Watermarked outputs
74 - Easy voice conversion script
75 - [Outperforms ElevenLabs](https://podonos.com/resembleai/chatterbox)
76
77 # Tips
78 - **General Use (TTS and Voice Agents):**
79 - The default settings (`exaggeration=0.5`, `cfg=0.5`) work well for most prompts.
80 - If the reference speaker has a fast speaking style, lowering `cfg` to around `0.3` can improve pacing.
81
82 - **Expressive or Dramatic Speech:**
83 - Try lower `cfg` values (e.g. `~0.3`) and increase `exaggeration` to around `0.7` or higher.
84 - Higher `exaggeration` tends to speed up speech; reducing `cfg` helps compensate with slower, more deliberate pacing.
85
86
87 ***Note:*** Ensure that the reference clip matches the specified language tag. Otherwise, language transfer outputs may inherit the accent of the reference clip’s language.
88 ***To mitigate this, set the CFG weight to 0.***
89
90
91 # Installation
92 ```
93 pip install chatterbox-tts
94 ```
95
96
97 # Usage
98 ```python
99 import torchaudio as ta
100 from chatterbox.tts import ChatterboxTTS
101
102 model = ChatterboxTTS.from_pretrained(device="cuda")
103
104 text = "Ezreal and Jinx teamed up with Ahri, Yasuo, and Teemo to take down the enemy's Nexus in an epic late-game pentakill."
105 wav = model.generate(text)
106 ta.save("test-1.wav", wav, model.sr)
107
108 # If you want to synthesize with a different voice, specify the audio prompt
109 AUDIO_PROMPT_PATH="YOUR_FILE.wav"
110 wav = model.generate(text, audio_prompt_path=AUDIO_PROMPT_PATH)
111 ta.save("test-2.wav", wav, model.sr)
112 ```
113
114 # Multilingual Quickstart
115 ```python
116 import torchaudio as ta
117 from chatterbox.mtl_tts import ChatterboxMultilingualTTS
118
119 multilingual_model = ChatterboxMultilingualTTS.from_pretrained(device="cuda")
120
121 french_text = "Bonjour, comment ça va? Ceci est le modèle de synthèse vocale multilingue Chatterbox, il prend en charge 23 langues."
122 wav_french = multilingual_model.generate(french_text, language_id="fr")
123 ta.save("test-french.wav", wav_french, model.sr)
124
125 chinese_text = "你好，今天天气真不错，希望你有一个愉快的周末。"
126 wav_chinese = multilingual_model.generate(chinese_text, language_id="zh")
127 ta.save("test-chinese.wav", wav_chinese, model.sr)
128 ```
129 See `example_tts.py` for more examples.
130
131
132 # Acknowledgements
133 - [Cosyvoice](https://github.com/FunAudioLLM/CosyVoice)
134 - [HiFT-GAN](https://github.com/yl4579/HiFTNet)
135 - [Llama 3](https://github.com/meta-llama/llama3)
136
137 # Built-in PerTh Watermarking for Responsible AI
138
139 Every audio file generated by Chatterbox includes [Resemble AI's Perth (Perceptual Threshold) Watermarker](https://github.com/resemble-ai/perth) - imperceptible neural watermarks that survive MP3 compression, audio editing, and common manipulations while maintaining nearly 100% detection accuracy.
140
141 # Disclaimer
142 Don't use this model to do bad things. Prompts are sourced from freely available data on the internet.