Speech Processing

Argumentation

We have investigated argumentation task using audio features. We have also developed a toolkit to foster research on this topic.

Clinical domain

We have defined a taxonomy of best practices for curating and maintaining clinical datasets for audio modality. We are currently evaluating interpretable speech techniques on clinical data (e.g., depression).

Interpretability

The development of interpretable audio features to address downstream task. An example is audio tokens, a discretization process to better analyze audio inputs.

Multimodality

We have explored text and audio modalities to assess their overall and individual contribution. We have mainly target argumentation tasks for now.