Section: New Results

Audio and speech content processing

Audio segmentation, speech recognition, motif discovery, audio mining

Audio motif discovery and spotting

Participants : Frédéric Bimbot, Nathan Souviraà-Labastie.

This work was performed in close collaboration with Emmanuel Vincent from Inria Nancy-Grand Est.

As an alternative to supervised approaches for multimedia content analysis, where predefined concepts are searched for in the data, we investigate content discovery approaches where knowledge emerge from the data. Following this general philosophy, we pursued work on motif discovery in audio contents.

Audio motif discovery is the task of finding out, without any prior knowledge, all pieces of signals that repeat, eventually allowing variability. The developed algorithms allows discovering and collecting occurrences of repeating patterns in the absence of prior acoustic and linguistic knowledge, or training material. When the audio pattern is determined in a user supervised fashion, the task becomes that of motif spotting.

Investigated in the context of SPORES (SPOtted Reference based Separation) [13] , audio motif spotting has been illustrated as a useful way to exploit redundancy in audio contents, for guided source separation purposes.

Mobile device for the assistance of users in potentially dangerous situations

Participants : Romain Lebarbenchon, Ewen Camberlein, Frédéric Bimbot.

The S-Pod project is a cooperative project between industry and academia aiming at the development of mobile systems for the detection of potentially dangerous situations in the immediate environment of a user, without requiring his/her active intervention.

In this context, the PANAMA research group has been involved in the design of algorithms for the analysis and monitoring of the acoustic scene around the user, yielding audio-based information which can be fused with other sensors (physiological, positional, etc.) in order to trigger an alarm (and subsequent appropriate measures) when needed.

The last phase of the project has been dedicated towards robustness improvement of audio scene analysis, with a particular focus on threat vs non-threat detection, on the basis of adaptive training scenarii. Knowledge and know-how transfer has been achieved for the hardware implementation of the designed methods and the efficient integration into an operational prototype.