We build speech technologies and resources for diverse languages and real-world conditions.

Our work spans:

• Speech recognition: robust ASR for different languages, dialects, and acoustic conditions.
• Speech synthesis: natural and expressive TTS, including prosody and model post-training.
• Speech data: datasets and tools for training and evaluating speech models.

Featured release

🥁 Darbuka Arabic — an Arabic speech dataset for ASR research and TTS data preparation, with a focus on Egyptian and Saudi speech.

Explore our repositories to see what we’re working on.