cpp-timbreID

An audio feature extraction library for the JUCE framework.
Porting of some of the objects from the Puredata external wbrent/timbreID [1].
A wrapper for aubio/aubio's onset detector is included.

A earlier version of this library was used for the paper On the challenges of embedded real-time audio classification which was submitted to the 25th International Conference on Digital Audio Effects (DAFx20in22). The code for the timbre classifier mentioned in the article is in /master/Demos/Demo-guitarTimbreClassifier
Check the JUCE projects in Demos/ to understand how to use the library.

At an early stage of development, the library was used for the following scientific papers:

• A Comparison of Deep Learning Inference Engines for Embedded Real-time Audio Classification [2]
• On the challenges of embedded real-time music information retrieval [3]

Build and usage

At the current stage, library source files (in src/) need to be compiled alongside the user's project. However, most of the code is header-only, requiring to add include to the include path of the project.
The library also includes a convenient wrapper for Aubio Onset, which requires to compile the library in libs/audio. To do so, run the script libs/build_dependencies.sh, which build a tested version of Aubio for both amd64 and Elk Audio OS arm64/aarch64.

References

1. W. Brent, A timbre analysis and classification toolkit for pure data, in Proc. 2010 Int. Computer Music Conference (ICMC 2010), New York, USA, 2010.
2. Stefani, D., Peroni, S., & Turchet, L. (2022, September). A Comparison of Deep Learning Inference Engines for Embedded Real-Time Audio Classification. Proceedings of the 25-Th Int. Conf. on Digital Audio Effects (DAFx20in22), 3, 256–263. Vienna, Austria.
3. Stefani, D., & Turchet, L. (2022, September). On the Challenges of Embedded Real-Time Music Information Retrieval. Proceedings of the 25-Th Int. Conf. on Digital Audio Effects (DAFx20in22), 3, 177–184. Vienna, Austria.

Domenico Stefani
domenico [at] unitn [dot] it