Chatterbox Turbo ONNX
KitsuMate mirror of the official
ResembleAI/chatterbox-turbo-ONNX
repository for reproducible Unity and ONNX Runtime downloads.
Contents
Each precision family under onnx/ contains the four pipeline graphs:
| Stage | File |
|---|---|
| Reference speech encoder | speech_encoder{suffix}.onnx |
| Text and speech-token embeddings | embed_tokens{suffix}.onnx |
| Autoregressive language model | language_model{suffix}.onnx |
| Conditional acoustic decoder | conditional_decoder{suffix}.onnx |
Supported suffixes are empty (FP32), _fp16, _q4, _q4f16, and
_quantized. All files are byte-for-byte copies of the official ONNX source
revision listed below. Each graph must remain beside its matching .onnx_data
file. The model produces
24 kHz English speech and supports the paralinguistic tags provided by Chatterbox
Turbo.
Reproducibility
- Official ONNX source revision:
d21799bd0354adb85e348b8a0442a8405110a2cf - Original implementation:
resemble-ai/chatterbox - Original model:
ResembleAI/chatterbox-turbo - File hashes and byte sizes:
SHA256SUMS
default_voice.wav is redistributed unchanged from
onnx-community/chatterbox-ONNX
revision 3cab09af388d3f02bba43443fce88c1f4525ac43, under the same MIT license. Its
SHA-256 is 3ebc531cdaba358a327099c1c4f0448026719957bcf4d8e9868767f227e02f4e.
License and attribution
Chatterbox and its model weights are Copyright (c) 2025 Resemble AI and are licensed under the MIT License. This repository preserves the original license and provides no additional model-weight license.