Chatterbox Original: BookAlive native ONNX conversion
First-party Nile Academy conversion of the official English Original checkpoint at
5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18. This is a derivative ONNX artifact,
not an ONNX release published by Resemble AI. The independent conversion uses publisher
source at 5de7a54aa4e5e2baadb0182dde554908b48b85c2 and safetensors only.
No community exporter or community checkpoint is included.
The native adapter supports reference cloning, emotion exaggeration, classifier-free
guidance and temperature. The official ten-step flow decoder is preserved; Turbo's one-step
decoder is not substituted. Publisher defaults are exaggeration 0.5, guidance 0.5 and
temperature 0.8. For a fast reference, guidance near 0.3 can help; stronger exaggeration may
increase speaking speed. Breath sounds depend on the model and reference. There is no
guaranteed breathing knob and no artificial breath insertion.
The pack contains explicit language-model KV caches, reference frontends and encoders,
conditioning, a shared flow-estimator graph run for ten Euler steps, a vocoder, tables and
tokenizer. Stochastic vocoder phase/noise are host inputs for reproducible parity checks.
build_metadata.json separates deterministic component checks from listening assessment.
MIT checkpoint/publisher code, with Apache-2.0 and MIT-derived acoustic-component notices
retained in the pack. Workstation development build; unsigned manifest; no human listening
review or commercial release qualification. CPU numerical checks do not establish GPU speed.
Use only reference recordings authorized by the speaker.
The native adapter does not implement the publisher's Perth watermark postprocessor.
Watermark/runtime/licence release review remains open for this unsigned development build.
Official implementation: https://github.com/resemble-ai/chatterbox
