All projects

Multimodal emotion analysis

Reading emotion from counselling audio and text together

TebahSoftMar 2023 – Jul 20235 mos

Context

I owned the emotion analysis module of a product that analyses counselling sessions.

Problem

Text alone loses the emotional signal carried by tone and intonation, so the analysis was hard to trust.

How I solved it

  1. Rather than scaling the text model, I went after the signal it never had — tone and intonation are absent from a transcript, so no amount of model size recovers them
  2. Designed an end-to-end inference structure that folds audio in as a second modality
  3. Adopted a fine-grained emotion taxonomy to capture subtler states
  4. Built the multimodal preprocessing pipeline and a real-time inference server on BentoML

Result

  • F1 score improved 10% over the text-only baseline

Impact

Emotional signals that text alone missed now reach the analysis.

Tech

PythonPyTorchStreamlitBentoML
NextCounselling summary and diarizationShipped to production