Section: New Results

Real-time Audio Sources Classification

Participants : Christophe Biernacki, Maxime Baelde.

Recent research on machine learning focuses on audio source identification in complex environments. They rely on extracting features from audio signals and use machine learning techniques to model the sound classes. However, such techniques are often not optimized for a real-time implementation and in multi-source conditions. It is proposed here a new real-time audio single-source classification method based on a dictionary of sound models (that can be extended to a multi-source setting). The sound spectrums are modeled with mixture models and form a dictionary. The classification is based on a comparison with all the elements of the dictionary by computing likelihoods and the best match is used as a result. It is found that this technique outperforms classic methods within a temporal horizon of 0.5s per decision (achieved 6errors on a database composed of 50 classes). This work has been now extended with success to the multi-sources classification case and also the computational load has been sufficiently reduced to reach the real time target (less than 50ms). This work has been presented to an international conference in Signal Processing [25] and also to a national conference [26]. A preprint is well advanced and should be submitted to an international journal at the end of 2017.

It is a joint work with Raphaël Greff, from the A-Volute company.