Ground control to Major Tom… CRIM’s state-of-the-art technology used to decipher air traffic control speech!

Ground control to Major Tom…
CRIM’s state-of-the-art technology
used to decipher air traffic control speech!

Summary : CRIM’s experts and researchers take part in various technological challenges and evaluations every year. These events are essential, both to compare CRIM’s work to that of other experts in the field and to accelerate innovation by discovering the most successful techniques through challenges where all participants process the same data.

In 2018, our experts in speech recognition and Natural Language Processing (NLP) took part in the Airbus Air Traffic Control Speech Recognition Scientific Challenge (ATC 2018). This challenge was organized by Airbus in order to find the best speech recognition and textual analysis technologies to fit their business needs.

The ATC 2018 Challenge

The challenge was based on a very specific type of audio signal: radio communications between aircrafts and traffic control operators. Two tasks were given to the participants:

  1. Transcribe what is being said during the communications into text (using standard speech recognition and speech-to-text transcription)
  2. Automatically detect “call signs” in the communications. Call signs are specific code words used to address aircrafts. Generally, a call sign contains the carrier's code followed by a few alpha-numeric characters. To complete this task, CRIM's experts used sequence labelling and pattern detection techniques pertaining to the field of NLP.

Did you know?
The challenge used actual recordings of conversations between the control tower at Toulouse airport and aircrafts landing or taking off from that airport. All conversations were recordings of real interactions and actual traffic at Toulouse Airport. 
Want to hear an example of the type of conversation that the participants had to analyze?
Click here!


Why participate?

This international competition was open to companies, research centers, universities and individuals. The many complexities pertaining to ATC speech made it a great opportunity to test the limits of CRIM’s state-of-the-art speech recognition software.

Specific challenges of ATC speech include:

  • a lot of background noise;
  • very specific and specialized vocabulary (for example the NATO phonetic alphabet used to spell names or codes - “Alpha, Bravo, Charlie, etc.”);
  • and the operators’ high speech-rate.


Albeit the many challenges our experts faced, CRIM obtained impressive results during this competition. Participants’ results were presented at a workshop in Toulouse in October 2018. Among about 20 participating teams, CRIM ranked 3rd in the overall challenge and 2nd for the call sign detection task, a very impressive achievement for a first participation in this type of challenge combining speech recognition and textual analysis!

Learn more about CRIM’s participation in technological challenges and how CRIM’s expertise has helped propel innovation forward in the field!


Recent news

  • UEAT et le CRIM élaborent un système de recommandations de plats grâce à l’IA

    «Le CRIM nous a fait faire un bond de trois ans.»
    R-D industrielle, accompagnement technologique et scientifique personnalisé : le CRIM propulse les entreprises québécoises par des interventions ciblées adaptées à leurs besoins.


Upcoming event

  • Gestion de l'innovation 2020 - Événement Les Affaires
    10 September 2020 8:30
    Présentation en ligne
    Le CRIM est fier d'être partenaire de la 7e édition de la conférence Gestion de l'innovation organisée par les Événement Les Affaires. Présentation en ligne.
  • Le CRIM prend aussi part à des grands projets de recherche nationaux et internationaux visant à répondre à d’import…
  • AQT RT @aqtech: [#LecturedujourAQT] Notre #membreAQT, @UEAT, ainsi que notre partenaire de l'écosystème techno, le @CRIM_ca élaborent un systèm…

Recent Publications

  • The Indigenous Languages Technology Project at NRC Canada: an empowerment-oriented approach to developing language software

  • On the use of the i-vector speech representation for instrumental quality measurement