Martijn Bartelds

Accepted at #ICLR2026! 🎉🇧🇷 Deep learning models often fail on specific subgroups. Group DRO was designed to help, but fails when group losses aren't comparable. This is common in speech. We introduce CTC-DRO: up to 47.1% lower worst-language errors in multilingual ASR 👇

✨Meet OLMoASR✨ By pairing our curated 1M-hour dataset with a powerful architecture, we've built open ASR models that achieve competitive performance with models like Whisper. We're open-sourcing data, code and models to help the community build more robust and transparent ASR.

Now that school is starting for lots of folks, it's time for a new release of Speech and Language Processing! Jim and I added all sorts of material for the August 2025 release! With slides to match! Check it out here: web.stanford.edu/~jurafsky/sl...

Big THANK YOU to the amazing #Interspeech2025 Organizing Committee! 💙

🎤 Odette Scharenborg, Catharine Oertel, Khiet Truong
💰 Martijn Bartelds
🌐 Dragoș Bălan
🗂️ Saskia Peters
🤝 Ginny Ruiter, Marie Louise Verhagen, Natascha Voskuijl

Congratulations!! That’s wonderful!! 🎉🍾

Congrats!!! 🎉

CTC-DRO can be applied to ASR with minimal computational costs, and offers the potential for reducing group disparities in other domains with similar challenges.

📄 Read our paper: arxiv.org/pdf/2502.017...
💻 Get the code: github.com/Bartelds/ctc...

The result:
📊 Worst-language error ↓ up to 47.1%
📊 Average error ↓ up to 32.9%

CTC-DRO works seamlessly with existing self-supervised speech models through ESPnet 🚀

We present CTC-DRO, which addresses the shortcomings of the group DRO objective by:
✅ Input length-matched batching to mitigate CTC’s scaling issues
✅ Smoothing the group weight update to prevent overemphasis on consistently high-loss groups

Why? Group DRO needs comparable training losses between languages. But in ASR, CTC-based losses vary due to differences in speech length, speakers, and acoustics. This creates spurious differences across language groups.

Result? Worse performance.

We need a new approach 🚀

CTC-based fine-tuning has been successful in multilingual ASR benchmarks but it doesn't fix language performance gaps. Group DRO could help by focusing on worst-performing languages, but it does not work ❌

🎙️ Speech recognition is great - if you speak the right language.

Our new @stanfordnlp.bsky.social paper introduces CTC-DRO, a training method that reduces worst-language errors by up to 47.1%.

Work w/ Ananjan, Moussa, @jurafsky.bsky.social, Tatsu Hashimoto and Karen Livescu.

Here’s how it works 🧵

I am excited to announce that I will join the University of Zurich as an assistant professor in August this year! I am looking for PhD students and postdocs starting from the fall.

My research interests include optimization, federated learning, machine learning, privacy, and unlearning.

📢 Join us for the Conversational AI Reading Group meeting on Thursday, January 16th, 11 AM-12 PM EST.
Martijn Bartelds will present "Improving Universal Access to Modern Speech Technology".
Details here: poonehmousavi.github.io/rg

Happy New Year everyone! Jim and I just put up our January 2025 release of Speech and Language Processing! Check it out here: web.stanford.edu/~jurafsky/sl...

Group picture of people in the Stanford NLP Group gathered in front of the shores of Lake Tahoe.

Natural Language Processing—artificial intelligence that uses human language—has been on a roll lately. You’ve probably noticed! So the Stanford NLP Group has been growing, and diversifying into lots of new topics, including agents, language model programs, and socially aware #NLP.

nlp.stanford.edu

Excited to announce the launch of our ML-SUPERB 2.0 challenge @interspeech.bsky.social 2025! Join us in pushing the boundaries of multilingual ASR and LID! 🚀

💻 multilingual.superbbenchmark.org

Hi speech people, super exciting news here!

We are running another "Multimodal information based speech (MISP)" Challenge at @interspeech.bsky.social

Participate!
Spread the word!

More info 👇
mispchallenge.github.io/mispchalleng...

made this thing, reply to be added
go.bsky.app/AKGJ82V

🙋‍♂️

Mentioning this post from @cjziems.bsky.social, listing some starter packs: bsky.app/profile/cjzi...

I've started putting together a starter pack with people working on Speech Technology and Speech Science: go.bsky.app/BQ7mbkA

(Self-)nominations welcome!

🙋‍♂️

I wanted to contribute to "Starter Pack Season" with one for Stanford NLP+HCI: go.bsky.app/VZBhuJ5

Here are some other great starter packs:

- CSS: go.bsky.app/GoEyD7d + go.bsky.app/CYmRvcK
- NLP: go.bsky.app/SngwGeS + go.bsky.app/JgneRQk
- HCI: go.bsky.app/p3TLwt
- Women in AI: go.bsky.app/LaGDpqg

👋