The client is a leading music tech company that offers its customers sound packs comprising one-shot samples and musical loops.
The client needed three AI classifiers to extract musical knowledge from audio files. The most important was a music instrument classifier. The instrument classifier should automatically identify the main instrument present in one-shots and loops among 100+ possible instruments. The client also required algorithms that should automatically detect the level of reverb and distortion present in an audio file.
We initially built prototype classifiers for instrument classification reverb and distortion detection in the Python programming language. We used machine learning and digital signal processing techniques to develop the AI, training it on a dataset of loops and one-shots we prepared. As the second step, we turned the prototypes into production-ready solutions refining the algorithms and porting the Python code to the C++ programming language.