Model reference · open weights
Chatterbox-Multilingual-zh-cmn is an open-weight audio or speech model from ResembleAI. Chatterbox-Multilingual-zh-cmn (BF16) weighs 3.2 GB; the smallest configuration that runs it is RTX 3060 12 GB.
Summary of the ResembleAI/Chatterbox-Multilingual-zh-cmn model card, 2026-10-01
What it is
| Released by | ResembleAI |
|---|---|
| Released | 2026-04-22 |
| VRAM | 3.2 GB for the weights |
What it runs on
| Card | One stream | Memory |
|---|---|---|
| RTX 3060 12 GB | fits | 11.6 GB |
| RTX 4060 Ti 16 GB | fits | 15.4 GB |
| RTX 3090 24 GB | fits | 23.4 GB |
| RTX 4090 24 GB | fits | 23.4 GB |
| RTX 5090 32 GB | fits | 31.0 GB |
| L40S 48 GB | fits | 44.0 GB |
| A100 80 GB | fits | 78.2 GB |
| H100 80 GB | fits | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | fits | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | fits | 107 GB |
| H200 141 GB | fits | 138 GB |
| B200 180 GB | fits | 176 GB |
From the model card
🎙️ Live demo: Try this model in the
ResembleAI/Chatterbox-Multilingual-TTS-zh-cmnSpace.
Chatterbox Multilingual: Mandarin Chinese is a dedicated single-language finetune in the Chatterbox Multilingual V3 Single Language Pack. It is optimized for Chinese Mandarin, with language- and region-specific behavior for expressive text-to-speech and voice cloning.
Use this model when you want tighter Mandarin Chinese quality control than the broad multilingual checkpoint. For a single model that covers all supported languages, use ResembleAI/chatterbox.
Try the hosted demo Space: ResembleAI/Chatterbox-Multilingual-TTS-zh-cmn.
t3_zh_cmn.safetensors: T3 state dict in safetensors format.s3gen_v3.pt / s3gen_v3.safetensors: V3 S3Gen speech decoder checkpoint.grapheme_mtl_merged_expanded_v1.json: multilingual tokenizer config.ve.pt: voice encoder checkpoint used by the demo.conds.pt: built-in reference voice conditioning used by the demo.zh-CMNzh88000t3_088000.pth.tar292float32(2454, 1024)(8194, 1024)2143990176 bytes4c51f16551e4f0a5e7bb88c58c4fc1f4ab2f0a47173665858553c8b76c900fe7This repository contains Chatterbox Multilingual V3 single-language assets used by the linked demo Space. The T3 checkpoint is loaded with multilingual vocabulary shape 2454 and S3 speech vocabulary shape 8194.
The demo combines these model-specific assets with the shared Chatterbox inference code and companion assets needed for end-to-end speech generation.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.
How it works