What's new
- Reference-audio conditioning and zero-shot voice cloning.
- Reusable conditioning through condition() and generate_conditioned().
- Upstream AuK parity corrections for reference-audio inference.
- Alignment with TinyTalk's Python 3.13, PyTorch 2.11 and Transformers 5.17 runtime.
- Simplified CPU inference configuration and removal of experimental optimization paths.
Validation
Qualified CPU reference-audio synthesis with the upstream-compatible inference path.
Full Changelog: v0.1.1...v0.2.0