Audio Foundry (previously CAT-Talk) is a secure, web-based platform that combines the latest speech-to-text transcription and speaker labeling (diarization) to produce accurate, speaker-labeled, time-stamped transcripts of audio and video files quickly, with summarization and theme extraction on top.
The work began as an answer to the administrative burden of patient note-taking. Evaluation has covered more than 40 hours of medical conversations between patients and providers, and the system handles complex conversations, noisy environments, and overlapping dialog, with the aim of making human review more efficient.
Part of the Kentucky Open Science platform: it runs over the Cresco mesh, with versioned storage for datasets, models, and code in GaiaKeep. See the full stack.
Two open-source models are integral to the system:
Diarization and transcription are combined into one time-stamped, speaker-labeled transcript with WhisperX, and uploaded files are scheduled as jobs so many run at once.
The transcription platform is available for experimental use. Request access through the information request form. A training video is available on YouTube.
M. A. Klusty, W. V. Logan, S. E. Armstrong, A. D. Mullen, C. N. Leach, K. Calvert, J. Talbert, V. K. C. Bumgardner. Toward Automated Clinical Transcriptions. AMIA Joint Summits on Translational Science Proceedings, 2025, pp. 235-241. PubMed