whatmodel.sh / tts / Chatterbox
Chatterbox
↗ share
+ contribute
Chatterbox is an open-weights text-to-speech model from Resemble AI, released May 2025. Its published size is 0.5B parameters. Licence: mit.
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of Chatterbox, out of 148 with a score, then more from Resemble AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 84 |  LEMAS-TTS (w/o prosody, multilingual_grl) | 50±6 | local | ||
| 85 |  MMS TTS Amharic | 50±16 | local | ||
| 86 | Chatterbox | 50±5 | api $25.00/M chars +local | ||
| 87 |  Parler-TTS Mini v0.1 | 49±14 | local | ||
| 88 |  Fish Speech 1.2 | 49±14 | local | ||
| more from Resemble AI | |||||
| 69 | Chatterbox Multilingual V3 | 54±10 | local | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
other tests4 not in a task · 91% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| EmergentTTS · 6-category preference | 24.5% ✎ | ▅▅▃▃█▅▅▃·▃·▅ beats 6/16 | 63% |
| EmergentTTS · 6-category WER | 13.1% ✎ | ▂▄··▂▄▄··█▄▆ beats 9/16 | 16% |
| CodecMOS-Accent VCTK voice clone · human speaker-similarity MOS (S-SPK-SIM) | 4.33 / 5 ✎ | ▄▄█·▄▄▄▄▄·▄█ beats 8/11 | 5.9% |
| CodecMOS-Accent VCTK voice clone · human naturalness MOS (S-NAT) | 4.23 / 5 ✎ | █··▄▄██▄·▄▄▄ beats 9/11 | 5.7% |
$ where to run it
Through an API
api · Replicate · $25.00/M chars · replicate.com ↗ · price sourceUSD per 1M input characters: Replicate's resemble-ai/chatterbox model page bills $0.025 per thousand input characters (x1000). Resemble's own pricing page lists no per-character TTS rate. Product: Replicate hosted model (official resemble-ai account). API model: resemble-ai/chatterbox. Settings: standard. Price basis: published per-unit billing, not hardware time. Exclusions: taxes. Currency: USD. Replicate official model page, observed 2026-09-25.
Other ways to run it
version
- Run it as published · needs ~2.9 GB, NVIDIA only · from github.com/resemble-ai (official)
pip install chatterbox-tts && python -c "import torchaudio as ta; from chatterbox.tts import ChatterboxTTS; m = ChatterboxTTS.from_pretrained(device='cuda'); ta.save('out.wav', m.generate('Hello world'), m.sr)"
Official weights and documentation: huggingface.co/ResembleAI ↗
$ how the score was worked out · EmergentTTS-Eval six-category… 79%, CodecMOS-Accent: listening-te… 12%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 EmergentTTS-Eval six-category leaderboard | 2 | 79% |
 CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1) | 2 | 12% |
| prior · pull toward the middle | 9% | |
| weighted average | = 49.5 |
$ published speed measurements · 1
Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.
| NVIDIA RTX 4090 | 2.004008 x-realtime | chatterbox-streaming fork · Linux; streaming example; first chunk latency 0.472s; 5.840s audio; 2.915s generation Quantization/dtype not specified in benchmark claim. Reciprocal of published rounded RTF 0.499; extra decimal places do not imply additional measurement precision. | source ↗ |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json