The idea
Every customer support call is "recorded for quality assurance" — but most of that data sits unused. Either companies hire third-party firms to manually review calls, or the recordings are never analysed at all.
This project automates the entire pipeline: transcription → summarisation → categorisation → insight extraction, turning every recorded call into structured, actionable data.
How it works
Whisper transcription
Audio recordings are transcribed to text using OpenAI's Whisper, handling multiple accents and audio quality levels.
GPT summarisation
Transcripts are processed by GPT-3.5 to generate concise summaries, key insights, and automatic call categorisation.
Insight extraction
Sentiment, topics, and action items are extracted automatically — surfacing patterns across thousands of calls.
Gradio interface
Clean web UI for uploading recordings and viewing results — deployed on Hugging Face Spaces.
Built with
Whisper AI
GPT-3.5
Python
Gradio
Hugging Face