We are building out state-of-the-art models, tools, and datasets to make speech technologies more inclusive for all dialects.
Research
Interspeech Conference Science Slam '25
We presented our work on making speech pathology more accessible, insights and statistics into our latest phonetic model, and a demo of our work-in-progress pronunciation platform.
University of Washington Research Symposium '25
We presented architectural challenges in building speech tools and models, working with non-lab settings and real-time data, and our mission as a research lab.
Models
We introduce a state-of-the-art phonetic transcription model, XLSR-English-01, which transcribes both standard and non-standard English speech, achieving a 60% improvement in feature error rate over other phonetic models, including Facebook's model, Facebook 60.
To develop truly inclusive speech technology, models should be able to accurately transcribe diverse speaker profiles, from speech impediments to non-native speech.
In light of this, we test our model on speech aphasia data (PSST) and non-native speech spanning from Italian and German (ISLE) to Mandarin (SpeechOcean) accented speech.
This model marks our second major update from our previous SoTA model (Koel-Labs-B0), and we aim to continuously improve upon phonetic transcription models in both their transcription quality and inference time.
Previews
XLSR-English-01, the state-of-the-art model for phonetic transcription, is available on Huggingface, alongside our other models, various datasets, and our IPA transcription leaderboard.
Phrase Segmentation, our demonstration tool for how phonetic transcription models will eventually interact with users in conjunction with language learning materials, is now available in closed beta.
A robust and open speech analysis tool, powered by our own models, to allow for real-time feedback to users about speech, based on an actor reference, not a restricted, arbitrary standard.
Early Access
We’re inviting a small group for early access to our research previews. Reserve your spot today.