Speech and Text

The Speech and Text team at CRIM offers various services for data processing and interpretation in fields such as real-time closed-captioning, exploitation of large datasets, multimedia indexing, information protection, audio signal analysis and segmentation, voice biometrics and natural language processing. 

Our expertise

Our team's expertise mainly covers the fields of automated speech recognition, speaker recognition, audio signal processing and  natural language processing (NLP). With all this knowledge, combined with an expertise in probabilistic methods and deep learning, the team is strategically equipped to contribute to scientific advancement and to respond to companies' specific needs. The team has developed numerous ground-breaking algorithms that have been adopted worldwide.

See all expertise

Profile of an expert

Gilles Boulianne
Gilles Boulianne
Researcher

At CRIM since 1998, Gilles Boulianne is interested in transducer-based search techniques aimed at accelerating system operation and/or increasing the size of the vocabulary of the CRIM speech recognition system.

Contact The Speech and Text Team

For any information, please contact
Hans Bherer, director.

514 840-1235, poste 4624

Recent news

  • CRIM and the NFB: The power of collaborative partnerships to drive technological and social innovation
    14/11/2018

    A current project aims to optimize CRIM's video description technology and extend its availability to a wider audience using the NFB's broadcasting platforms.

    +

Upcoming event

  • Intelligence d'affaires et analytique 3e édition - Événement les Affaires
    14 November 2018 8:30
    Centre-Ville, Montréal
    Le CRIM est fier d'être partenaire et de participer à la 3e édition de la conférence Intelligence d'affaires et analytique organisée par les Événements Les Affaires, le 14 novembre 2018.
    +

Recent Publications

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

    +
  • Checking Sequence Generation for Symbolic Input/Output FSMs by Constraint Solving

    +