Alberto Abad

A groundbreaking multimodal dataset that captures the complexities of speech and non-verbal communication across diverse medical conditions, paving the way for more nuanced AI analyses.

Achieving a 66% reduction in parameters without sacrificing ASR performance opens new avenues for efficient model design.

European Portuguese speakers finally get a seat at the table: FalAR, a massive new speech corpus, closes the gap with Brazilian Portuguese and boosts ASR performance by 14%.

You can protect patient privacy and still detect Parkinson's from speech, but only if you choose the right anonymization method.

Model merging can beat full fine-tuning for multi-domain ASR, offering a scalable path to adapt large speech models without prohibitive computational costs.