Release Date: May 13, 2026
AudioMuse AI v1.1.4 improves Lyrics Analysis speed and reliability with smarter recognition of instrumental and voiced tracks. Instrumental songs can now be skipped earlier, avoiding unnecessary ASR processing, while voiced tracks are less likely to be incorrectly discarded thanks to improvements based on the Musicnn classifier and VAD fine-tuning.
For users who do not need speech recognition at all, the Setup Wizard now also allows ASR to be completely disabled by setting LYRICS_ASR_ENABLE to false.
We also migrated the lyrics models from PyTorch to ONNX, improving stability across updates and different lyrics models while keeping full compatibility with previous analyses. ONNX continues to be the key technology that allows AudioMuse AI to run consistently across multiple architectures, including ARM devices.
Alongside these improvements, v1.1.4 also includes multiple fixes and general optimizations across the project.
What's Changed
- Setup wizard advanced parameter default value + index.html fix - #506 by @NeptuneHub in #513
- Fixed dashboard timezone - #499 by @NeptuneHub in #514
- Backup Restore improvement by @NeptuneHub in #517
- Lyrics - improve vad recognition and language detection by @NeptuneHub in #518
- updated apidocs by @NeptuneHub in #519
- Pytorch To Onnx + Qwen ASR by @NeptuneHub in #522
- LYRICS_ASR_ENABLE + Log cleaning + dockerfile cleaning by @NeptuneHub in #529
- mood check improvement in lyrics analysis by @NeptuneHub in #530
- Worker RAM saving during analysis fix by @NeptuneHub in #531
Full Changelog: v1.1.3...v1.1.4