v1.3.0

We are excited to announce the release of v1.3.0! This update includes several enhancements and usability improvements.

Highlights

UTMOSv2 now supports audio that is already in memory! This would be useful for many use cases, including employing UTMOSv2 as a reward model or integrating UTMOSv2 into custom pipelines.
See the corresponding issue and PR for more details:

• #81
• #82

What's Changed

• Add py.typed marker file for PEP 561 compliance by @kAIto47802 in #68
• Relax torch version requirement by @kAIto47802 in #72
• remove audio file whitelist, fallback to numpy load if audio can't be read by librosa backend by @tkanarsky in #75
• Update ci conditions by @kAIto47802 in #76
• Support Python 3.13 by @kAIto47802 in #77
• Update type annotations by @kAIto47802 in #80
• Support audio that is already in memory by @kAIto47802 in #82
• Refactor core modules by @kAIto47802 in #83
• Update README by @kAIto47802 in #84
• Add test_predict by @kAIto47802 in #85
• Add sr argument for the user input by @kAIto47802 in #86
• Update README.md by @kAIto47802 in #87
• Update README.md by @kAIto47802 in #88
• Fix device mismatch in torchaudio Resample when input is on CUDA by @kAIto47802 in #90
• Introduce uv by @kAIto47802 in #91
• Bump up to v1.3.0 by @kAIto47802 in #92

New Contributors

• @tkanarsky made their first contribution in #75

Full Changelog: v1.2.1...v1.3.0