Chatterbox Multilingual
An open-weights, 500M-parameter zero-shot TTS model from Resemble AI covering 23+ languages, released under the MIT licence.
- Weights
- Downloadable
- API
- Yes
- Commercial use
- Permitted subject to conditions
It clones a voice from a short reference clip without retraining and exposes emotion/intensity exaggeration controls. Every generated file carries Resemble's Perth neural watermark, with an open-source detector.
Licence is MIT, which permits commercial use, modification and redistribution with attribution and no copyleft.
- Output is watermarked. Note that all output is watermarked with Resemble's Perth watermarker and the detector is open source, so generated audio is identifiable
Specification
- What it takes in
- text, audio
- What it makes
- audio
- Controls
- voice-cloning, emotion-control, style-prompting
- Weights
- Downloadable
- API
- Yes
- Pricing
- Free to self-host under the MIT licence; compute cost only. Resemble also sells hosted variants (Turbo, Multilingual, Nano, Flash), but its current pricing page covers deepfake detection products rather than TTS, so hosted TTS rates could not be verified.
- Released or updated
- 2025-09 (Chatterbox Multilingual announcement, per third-party report)
- Languages
- 23, not individually listed by the vendor
Strengths
- MIT licence - genuinely permissive for commercial video work
- 23+ languages with zero-shot cloning from a short clip
- Emotion exaggeration control
- Self-hostable, so no per-character cost and no data leaving your infrastructure
- Benchmarked favourably against closed systems in vendor-run comparisons
Weaknesses
- Built-in watermarking cannot be assumed removable and marks output as synthetic
- Sample rate not stated on the model card
- No consent verification whatsoever on cloning
- Requires GPU infrastructure and engineering effort
- Vendor's commercial attention has shifted to detection products, which raises questions about long-term maintenance
Best for
- self-hosted multilingual VO
- on-prem or air-gapped production
- teams needing licence certainty
Where the sources disagree
Recorded rather than resolved. Picking the tidier number would hide the fact that the vendor's own pages do not agree.
- Language count is given as '23 languages' in Resemble's announcement and '23+' on the model card, which lists 23 named languages. Recorded as 23.
- Resemble's own comparative claim that Chatterbox is 'consistently preferred' over ElevenLabs is vendor-run evaluation, not independent benchmarking.
What to understand before using it
Explainers naming the mechanisms this model's own documentation mentions.
See it from the vendor
Shotcraft hosts no sample output. Judging a model means watching its own showreel with your own eyes, and a clip re-hosted here would be a clip we chose.
Nearby in the directory
Azure AI Speech text-to-speech (Neural, Neural HD, MAI voices)
Microsoft's enterprise speech synthesis service, with 200+ prebuilt neural voices across 80+ locales plus tiers for HD voices, multilingual voices, mu
Voice and speech
Sonic 3.5
Cartesia's current TTS model, documented at sub-90ms first-byte latency across 42 languages. It is tuned for conversational delivery and for reading a
Voice and speech
XTTS-v2
An open-weights multilingual TTS and voice cloning model covering 17 languages at 24kHz, still widely used for local voice cloning. Coqui Inc. shut do
Voice and speech