Caleb Rascòn (Caleb Rascon) and I propose Gray Jedi MVDR Post-filtering, which uses target (the light side) and interference (the dark side) to improve post-filtering using a low-complexity nn. Paper: arxiv.org/abs/2309.05057 Demo: youtube.com/watch?v=8Gr-xv…

Preprint from my student William Aris that describes how lightweight sound source localization can propose a region of interest for visual face detection. Results show a reduction of computations compared to parallel multimodal fusion approaches. Paper: lnkd.in/e3-TWxmB

Here is a demo of the work from my student William Aris, that shows how an audio-video camera can provide a specific region of interest for face detection, and reduce the computing load. youtu.be/EmVnvBFdjro

In this work we tried to do e2e meeting transcription with a "Whisper-inspired" approach using autoregressive enc/dec (AED) ASR. Modular approaches have their advantages IMO (more flexibility) but fully e2e is also intriguing. Our approach is quasi e2e as we use clustering.

New workshop announcement 📢📢 We are excited to announce our IEEE ICASSP 2024 workshop, Explainable AI for Speech and Audio! We have two tracks for papers with deadlines January 20, and February 20. The details on how to submit: xai-sa-workshop.github.io/web/Call%20for…

When I review a paper and I get lost in the maths of the author, I always wonder if that’s because the paper is confusing (and I should review accordingly) or because I’m just stupid.

If you run out of greek letters for your variables, it means you have too many equations already in your 5-page paper.

Done reviewing 15 ICASSP papers - this took up most of my evenings this week. I'll have to learn to say no.

I am honored to have been elevated to IEEE Fellow "for contributions to multi-source speech and audio processing". I've been blessed w/ fantastic collaborators over the years, from advisors & lab mates to colleagues & interns, & am greatly thankful to them all🤗 IEEE Signal Processing Society IEEE

Just received this new book! Looking forward to reading it during the holidays once the semester is over!

🧠🗣️ Curious about using speech as a biomarker for early detection of neurodegenerative disorders? 🔬 We have Exciting Opportunities with 🎓 Open positions for #Postdocs and #PhD students in cutting-edge #DeepLearning #research. 📌 More info: drive.google.com/file/d/1QqiRgF…

SpeechBrain 1.0 is out!

Can Elsevier provide us with insights regarding its peer-review process and academic integrity? This paper was flagged yesterday, and the compromising text was quickly removed from the introduction. We can still see this part that clearly comes from ChatGPT. #chatgpt #elsevier

Multichannel sound separation is a challenging task with supervised learning, as it usually requires a multichannel dataset with isolated sound segments for training. This work carried out by Jacob Kealey (Université de Sherbrooke), John Hershey (Google) and François Grondin

I like honest reviews when a paper gets rejected, and tell my students to seize this as a opportunity to improve. However comments such as “There are people in Taobao selling such demo” do not achieve anything constructive, and sends a sad message to new grad students. #iros

One nice thing about working on a campus outside the city: I get to enjoy these moments with the dogs between two zoom calls.

Our team just won first place at the XPrize competition, which comes with a 5M$ prize. My student Jacob Kealey and I were in charge of the microphone array system to classify, localize and track birds in the wild!

Bel article ce matin dans La Presse pour présenter les travaux et l’entreprise de mon étudiant au doctorat Simon Michaud qui conçoit un système de désherbage robotisé pour réduire l’utilisation d’herbicide et palier à la pénurie de main d’œuvre. lapresse.ca/actualites/env…

I’m leaving Twitter and switching to BlueSky. You can follow me there: francoisgrondin