speech2speech summarizer : a tool for summarizing spoken language into concise audio summaries.
Takes spoken audio, transcribes it, summarises it, and reads the summary back to you. Built for things like meeting recordings and lecture transcriptions where you want the gist without re-listening to the whole thing.
How it works
- Audio analysis — uses librosa to extract features like pitch, volume, and speaking rate. Segments where the speaker emphasises something (louder, slower, pitch changes) get flagged as more important
- Summarisation — runs both extractive (TextRank) and abstractive (fine-tuned T5 transformer) approaches. The emphasis markers feed into both to weight important segments higher
- Text-to-speech — converts the summary back to audio, preserving some of the original emphasis and intonation
What I found
I ran a comparison study between the two summarisation approaches. Extractive summaries preserved the original phrasing but felt disjointed. Abstractive summaries were more coherent but sometimes missed nuance. A hybrid of both worked best.