Finite-state transducers

Finite-state transducers have allowed us to represent and treat several problems while guaranteeing optimal results and speed.

This meta-expertise has propelled CRIM to the forefront of high-technology since the end of the 1990’s. Indeed, any expertise at CRIM can benefit from it; whether it is speech recognition, language modeling, pronunciations or acoustic models. In turn, these advantages extend to a never ending list of additional problems such as conversion into graphemes, generation of  N best hypothesis, indexing and searching,  error rate computation, etc. This meta-expertise relies on our powerful and stable transducer-handling software library.

Related technologies: Recognition engine, audio content indexing, closed-captioning, text/audio alignment, pronunciation/password verification 

 

Teams

Recent news

  • Best wishes for the holidays!
    18/12/2019

    Happy Holidays from all of us at CRIM!

    +

Upcoming event

  • Transformation numérique - Secteur public - Événement Les Affaires
    29 January 2020 7:00
    DoubleTree par Hilton, Montréal
    Le CRIM fier partenaire de la 1ère édition sur la Transformation numérique - Secteur public organisée par les Événement Les Affaires, qui se tiendra les 29 janvier 2020 au DoubleTree par Hilton, Montréal.
    +

Recent Publications

  • On the use of Fisher Vector Encoding for Voice Spoofing Detection

    +
  • Deep learning speech recognition for low-resource languages

    +