Concordia researchers develop novel approach to help Large Language Models learn from speech
Reports and Proceedings
Updates every hour. Last Updated: 15-Dec-2025 12:11 ET (15-Dec-2025 17:11 GMT/UTC)
Concordia University researchers unveiled a new audio-tokenization method, FocalCodec, that compresses speech into compact tokens while preserving meaning and quality. Concordia University
By using binary spherical quantization and focal modulation, FocalCodec dramatically reduces bitrate — making speech easier for large language models to process. Concordia University
In listening tests with 33 participants, reconstructed speech was judged nearly indistinguishable from original recordings, showing the new method keeps natural sound.
New research shows that three-quarters of the world are not getting enough omega-3. A new study finds that 76 per cent of people worldwide are not meeting recommended intakes of EPA and DHA, revealing a significant global public health gap.
The research is the first global review of national and international omega-3 intake recommendations across all life stages for generally healthy people.
ITHACA, N.Y. – A new study of two ponds, Texas Hollow Pond and Mud Pond in Central New York, reveals surprising and complex mechanisms for how carbon dioxide and methane in ponds are built up, stored and released.