SawtifAI is an AI-native audio and video intelligence platform, being built as an independent product in the IoTOcean portfolio. It is aimed at journalists, podcasters, educators, researchers and teams who work with recorded speech, and at anyone who needs to make audio and video understandable in more than one language.
The problem
Recorded conversations hold a lot of value that is hard to reach. Finding one moment in a two-hour interview means scrubbing through it by ear. Knowing who said what, and how they said it, takes careful listening. Making the same content available in another language usually means starting again with a new team and a new recording.
How SawtifAI approaches it
SawtifAI is designed to turn recordings into something you can read, search and navigate. Audio is planned to be extracted from video and transcribed with AI, with speakers separated and identified so the transcript shows who is talking. The transcript, the speakers and key events are planned to sit on a shared timeline, synchronised with playback, so clicking a line jumps straight to that moment.
Beyond transcription, SawtifAI is exploring analysis of vocal expression and emotion, multilingual translation and localisation, and AI-assisted dubbing that keeps each speaker’s identity, timing and vocal expression where that is technically possible. These are the hardest parts of the product, and they are treated as research and development work, not as promises.
The technology direction is a React and TypeScript interface on top of a Python and FastAPI backend, with FFmpeg for media processing, asynchronous processing for long recordings, and integrations with AI providers.
What is being built now
SawtifAI is in development, and none of its capabilities is production-ready. The work in progress covers:
- audio extraction from video;
- AI transcription;
- speaker diarisation and identification;
- interactive synchronised playback with transcript, speaker and event timelines;
- click-to-seek navigation;
- multilingual translation and localisation.
Vocal-expression and emotion analysis and AI-assisted dubbing are planned for later and are still being explored. Progress will be shared in the IoTOcean Journal. If you work with interviews, podcasts, educational media, meetings or video localisation and want to follow the work, get in touch.
