🎙️ Multilingual Speech Emotion Recognition

Detects emotion in Sinhala, Tamil, and English speech.

Language

Select the language spoken — affects normalization

Inference Mode

ensemble is most robust | gemaps is fastest | fusion is highest accuracy on English/Tamil


Emotions: Neutral · Happy · Sad · Angry · Fear

Modes:

  • fusion — Whisper-tiny encoder + eGeMAPS (best on English & Tamil)
  • gemaps — 88 acoustic features only, language-agnostic, ~50ms
  • ensemble — 60% fusion + 40% gemaps (recommended for Sinhala)