Skip to main content

SpeechBrain 1.0: Concordia's contribution to the future of AI conversations

The community-based resource supports a wide range of conversational AI applications — and it’s accessible to all
April 8, 2024
Side by side images of two men: the first with blonde hair, brown glasses and a beard and the other man with brown hair wearing a grey button-up shirt.
Founders Titouan Parcollet and Mirco Ravanelli

The open-source, all-in-one conversational artificial intelligence (AI) toolkit SpeechBrain 1.0 has now officially launched.

The new community resource was co-founded by Mirco Ravanelli, assistant professor in Concordia’s Department of Computer Science and Software Engineering, and Titouan Parcollet, research scientist based at the University of Cambridge. SpeechBrain is the result of an ongoing collaborative effort by a large international network of developers.

It offers a variety of functionalities, addressing a wide range of applications such as speech recognition, language modelling, speech synthesis and dialogue.

Project leader Ravanelli notes that while there are other systems out there that people can use to address their specific needs — as a kind of black box, SpeechBrain is about understanding, and sharing, how this technology works.

Accessible, valuable information

This project began five years ago when Ravanelli and Parcollet met through the Montreal-based academic lab Mila.

Founded on their shared belief that technology ought to be accessible to all, SpeechBrain is not a startup, nor is it an academic endeavour. Volunteering their time to this resource community, Ravanelli notes their ability to be unbiased, “We just share — whatever we think is valuable.”

The community-driven project has exceeded 200,000 downloads a month and now boasts hundreds of contributors worldwide, each bringing their specific expertise to the platform, available to users free of charge.

Learning through doing

Concordia has played an important role in building SpeechBrain 1.0 via Ravanelli’s lab and his students. “They are all part of the core development team; this is how the project started and they continue to contribute their knowledge,” he shares.

Some contributions also occur within the context of a course offered by Ravanelli every winter term — Conversational AI (COMP 499/691). The course is open to undergraduate and graduate students who have completed the requisite machine learning course. It offers an opportunity to learn about technologies such as speech recognition and large language models (LLM) and includes hands-on experience with SpeechBrain.

Recipes for success

While open to all, SpeechBrain is specifically intended for students, researchers and entrepreneurs with a background in computer science who are interested in knowing how these technologies work and in developing their own AI technologies.

Each set of instructions available on the platform is referred to as a recipe — step by step actions required to address a specific problem and achieve a desired outcome. Each so-called recipe allows users to replicate models from scratch and hopefully “cook up” something new that might then be shared with the SpeechBrain community.

“People are accustomed to using this technology,” Ravanelli says. “But we want people to understand how to build this technology.”

Visit the Concordia co-founded SpeechBrain 1.0 website for further details, tutorials, quick installation and of course, the code!

Back to top

© Concordia University