Portfolio
PRAXIS is a research group first and an events organization second. These projects define the intellectual agenda. Members join an existing project or propose their own at the monthly pitch session.
Extracting acoustic and semantic biomarkers of anhedonia from spontaneous speech using ClinicalWhisper. By crossbenching anhedonia against computational phenotypes, the next step is integrating large language models with reinforcement learning to model reward-prediction errors in vocal flat affect.
A speech model adapted for clinical interview audio, and the tool the group is furthest along on building. A preprint was posted to bioRxiv in June 2026; the work is being prepared for peer-reviewed submission alongside continued validation.
Zhou, C., Wu, M., Xiang, Y., & Itti, L. (2026). Cross-modal benchmarking of acoustic prosody and ventral striatal BOLD for depression-related anhedonia classification: A pre-registered study with the ClinicalWhisper pipeline. bioRxiv. doi:10.64898/2026.06.08.728970
A parallel analysis track benchmarking computational phenotypes against functional imaging data, run alongside the voice work.
Open-source computational psychiatry resources and reference implementations.
RepositoryThe twenty-paper undergraduate curriculum: foundations of RL, active inference, and computational models of mood.
RepositoryThe code behind ClinicalWhisper — the speech model adapted for clinical interview audio.