Emotion-Aware Bot

A prototype that asks whether digital conversations could listen the way a person does: with eyes and ears, not just keywords.

How it works

• DeepFace detects facial emotion from webcam frames.
• Vosk transcribes speech locally — never leaves the machine.
• VADER scores the text sentiment, and compares it to the facial signal for alignment.
• Gemma 3B via LM Studio generates a reply with both signals as context.
• pyttsx3 speaks the reply back, offline.

Why it matters

A clinical trial showed therapy bots can reduce depression and anxiety with text alone. The open question is whether emotional perception — catching the mismatch between “I’m fine” and a visibly tense face — adds enough signal to change how the bot responds. The prototype is a concrete experiment on that question, local-first so the sensitive data never leaves the device.