François Portet

Finetuning speech recognition models can either amplify or erase critical speaker group information, depending on the training focus and fairness interventions.

LLMs that excel at evolutionary search aren't just smart zero-shot reasoners; they're masters of "local refinement," making small, consistent improvements without wandering off into the semantic wilderness.

Bias against certain speaker groups is embedded in self-supervised speech models from the very first layers, complicating efforts to achieve fairness in speech recognition tasks.