Our services

Sound as Data

Process audio data to provide personalized listening experiences

The transformation of listening habits and the rapid growth of audio markets have led to new requirements for broadcasters, as well as increased expectations from listeners. How can these be met to remain one step ahead?

These offers are based on IRCAM’s audio descriptors:

  • “low-level” descriptors, involving roughly 4,000 coefficients that are calculated using complex scientific methods.
  • “high level” descriptors, which can result from algorithms or machine learning, that describe the music being listened to using simple parameters (meta data).


These audio descriptors can be used to automatically extract information from music recordings using different modules:


Estimation of genre, mood and instrumentation classes


Estimation of tempo, meter, rhythmic complexity, percussiveness, temporal positioning of beats and downbeats


Estimation of tonality


Estimation of the succession of music chords


Estimation of the temporal structure of an audio file


Sound summary of an audio track