Release Date: October 09, 2025
In AudioMuse-AI v0.7.0-beta, we’ve made a major architectural change by replacing TensorFlow with ONNX.
This transition aims to provide more consistent and standardized results across different CPUs.
During testing, we observed that the TensorFlow-based implementation could produce slightly different analysis values between AudioMuse-AI releases, even on the same machine. This was likely due to dependencies with unfixed versions. By adopting ONNX, we aim to ensure more stable and reproducible results across both software versions and hardware configurations.
Additionally, this update should expands compatibility to support a wider range of CPU models.
We also decided to moving forward and remove the support of Essentia legacy contaienr version.
⚠️ Important:
If you update to this version, we strongly recommend cleaning your database and re-running the analysis.
- You can either delete and redeploy the database and its volume, or
- Connect to the database and truncate the tables.
Before performing these operations, please make sure to back up your database.
Highlights
- Removed Tensorflow, added ONNX
- All .pb model transformed in .onnx
- Removed Essentia legacy container image
What's Changed
- Onnx -> main by @NeptuneHub in #100
Full Changelog: v2.0.0-model...v0.7.0-beta