Audio analytics

Our team encompasses a set of skills dedicated to the extraction of information from an audio signal. 

An audio recording often contains more than one speaker. The process of diarisation partitions the audio signal into speaker segments, even when speaker are not known beforehand. Keyword detection consists in identifying the presence of certain words within audio recordings, whereas topic detection determines if a given subject has been discussed during a conversation. When the task is to accurately determine the position of each word or phoneme of a recording, we  speak of text/audio synchronization. We have also developed algorithms aimed at detecting a person’s emotions from an audio recording. It then becomes possible, in an automated system, to modify the system’s behavior according to the user’s reaction in order to adequately satisfy his interlocutor.

Related technologies: keyword detection, audio content indexing, emotion detection 

 

Teams

Releases

Recent news

  • The Joy of Reading, by De Marque
    18/04/2018

    Before the web existed, we worked a lot in schools and were motivated to offer content in digital form, specifically francophone and Quebecois content, because what existed at the time was mainly in English...

    +

Upcoming event

  • R-D Seminar - Towards coherent, fluent and context-appropriate Natural Language Generation systems
    24 April 2018 11:00
    CRIM (405, avenue Ogilvy, bureau 101, Montréal)
    A presentation by Jad Kabbara, scholarship student within the Speech and Text team at CRIM, and Ph.D. candidate at the School of Computer Science at McGill University.
    +

Recent Publications

  • Towards Automatic Feature Extraction for Activity Recognition from Wearable Sensors: A Deep Learning Approach

    +
  • État des lieux des technologies web

    +