ACE : the signature of a class of sounds

The core of ORELIA's technological solution is the concept of Audio Content Extractor (ACE). An ACE is a classification model that can identifies the contents of a sound from his sound signature.


An ACE is based on a statistical modeling of characteristic parameters of the spectral envelope signals of the same semantic class.


This model is obtained by statistical learning. Each sounds class has its own ACE, that is to say, its own specific model.

This approach brings several benefits :

The recognition of a sound event does not require the storage of the learning database because all information is contained in the sound signature.

The ACEs obtained are generic: sensitivity is similar to the human ear especially in terms of reverberation and distance from the source


Additional information