Views
No views yet
onnx-community/chatterbox-onnx
(Apache-2.0) via the vconnx export pipeline. Only the VC path is present —
the LLM / tokenizer components are not included because the VC pipeline does
not require them.resemble-ai/chatterbox).| File | Size | Variant | Notes |
|---|---|---|---|
speech_encoder.onnx + .onnx_data | ~565 MB | fp32 | From onnx-community upstream |
speech_encoder_q8.onnx | ~216 MB | INT8 | Dynamic quantization (MatMul only) |
conditional_decoder.onnx + .onnx_data | ~516 MB | fp32 | From onnx-community upstream |
conditional_decoder_q8.onnx | ~252 MB | INT8 | Dynamic quantization (MatMul only, If-subgraph excluded) |
speech_encoder_q8.onnx:onnxruntime.quantization.quantize_dynamic (MatMul ops only).Gemm nodes in the S3 VQ codebook (project_down) are excluded from
quantization because quantize_dynamic decomposes Gemm(transB=1) without
transposing the weight — a known ORT preprocessing bug that corrupts the
resulting MatMul node. A pre-quantization patch transposes the weight and
rewrites those nodes as MatMul+Add before the quantizer runs.conditional_decoder_q8.onnx:onnxruntime.quantization.quantize_dynamic (MatMul ops only).If nodes with subgraph nodes; these subgraph nodes
are excluded from quantization to avoid a hang in ORT's session initializer.faster-whisper base.en on the vconnx reference demo clip
(10.7 s, source.wav):| Variant | WER | Size (MB) |
|---|---|---|
| fp32 | 8% | 1081 |
| INT8 | 8% | 468 |
chatterbox engine with quantized=True:1from vconnx import VoiceCloner
2
3cloner = VoiceCloner(engine="chatterbox", quantized=True)
4result = cloner.clone_voice("source.wav", "reference.wav", "output.wav")