# Enhanced voice cloning resources

Model: ResembleAI/chatterbox-turbo-ONNX
Revision: d21799bd0354adb85e348b8a0442a8405110a2cf
Model and ONNX export: MIT, as declared in the upstream model card.
Source: https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/tree/d21799bd0354adb85e348b8a0442a8405110a2cf
License source: https://github.com/resemble-ai/chatterbox/blob/5de7a54aa4e5e2baadb0182dde554908b48b85c2/LICENSE

The browser engine uses the repository's fixed Transformers.js 4.3.0 (Apache-2.0) and ONNX Runtime 1.31.0-dev.20260914-8d85527a0 (MIT), with their existing same-origin vendors and notices. Model graphs, external weights, configuration and tokenizer are mirrored at /static/models/voice-clone/enhanced/d21799bd0354adb85e348b8a0442a8405110a2cf/. Resource metadata includes exact size and SHA-256; large mirror files are not committed.

The selected precision is FP16 for embed_tokens and Q4 for speech_encoder, language_model and conditional_decoder. Generation is WebGPU only. No Perth post-processing is applied by this integration; do not describe its output as watermarked. Voice cloning requires the speaker's lawful consent.