As healthcare services increasingly move online, mental health therapy is also shifting to remote delivery. While video and audio platforms enable virtual sessions, subtle emotional cues often go unnoticed compared to in-person interactions. This paper presents a web-based therapy platform that supports secure audio/video communication, incorporating real-time speech-to-text transcription and speech emotion recognition to enhance therapist understanding during and after sessions. The platform segments sessions into multiple analysis segments, each capturing the transcript and detected emotion. For emotion recognition, a publicly available Hugging Face model based on Wav2Vec 2.0 is utilized, classifying emotions from raw audio. Testing showed reliable detection of emotions like angry, sad, and neutral, though audio quality significantly impacted results. The system was deployed using LocalTunnel due to cloud limitations. Overall, this platform demonstrates how integrating AI tools into online therapy can provide valuable emotional insights for therapists.
Kabir Kumar (Mon,) studied this question.