Speech and Language Processing

Speech and Language Processing

Voxtral

We present Voxtral Mini and Voxtral Small, two multimodal audio chat models. Voxtral is trained to comprehend both spoken audio and text documents, achieving state-of-the-art performance across a diverse range of audio benchmarks, while preserving strong text capabilities. Voxtral Small outperforms a number of closed-source models, while being small enough to run locally. A 32K context window enables the model to handle audio files up to 40 minutes in duration and long multi-turn conversations. We also contribute three benchmarks for evaluating speech understanding models on knowledge and trivia. Both Voxtral models are released under Apache 2.0 license.

New neuroscience research upends traditional cognitive models of reading

A new study finds that the left posterior inferior frontal cortex activates within 100 milliseconds during reading, playing a critical, early role in turning text into speech, challenging traditional models that assumed a slower, step-by-step process.

Dog–human vocal interactions match dogs’ sensory-motor tuning

Human-to-pet communication requires speech processing by the animal and adjustments of the human speaking rate to match their pet’s receptive abilities. This study shows that dogs and humans share similar but not identical speech processing mechanisms and that dog-human vocal interactions match dogs’ sensory-motor tuning.

Neural Dynamics lab |