Chatterbox
A solid pick for self-hosting and full data control.
Overview
Chatterbox, from Resemble AI, is an open-weight 500M-parameter TTS model built on a Llama 3 backbone. It covers 23 languages with voice cloning and an emotion-exaggeration control, plus six dedicated single-language finetunes (Chinese, Spanish, Portuguese, Hindi) for tighter quality. It ships under an MIT tag, but commercial-use terms aren't fully spelled out, so teams should double-check before shipping a product. Self-hosted, it runs on a single laptop-class machine and keeps every voice sample on your own infrastructure, regardless of the vendor's US base.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Voice cloning for products
- Multilingual content dubbing
- Voice-based AI agents
- Local audio prototyping
- Video content localization
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 0.3 GB | 2 GB | Any recent PC/Mac |
| Q8 · balanced | 0.5 GB | 3 GB | Any recent PC/Mac |
| FP16 · max quality | 1 GB | 3 GB | Any recent PC/Mac |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve ResembleAI/chatterbox \
--max-model-len 32768 \
--tensor-parallel-size 1 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |