Show HN: Trelis Tiron – open-weights multi-speaker meeting transcription
Category: ai-ml
Tags: speech-recognition, speaker-diarization, whisper, transcription, meeting
Score: 6.7/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 7)
Tiron is an open-weights multi-speaker meeting transcription model that jointly transcribes and attributes speech to speakers in a single decoding pass, using a Whisper large-v3 architecture with extended tokens for up to 8 speakers per 30-second window. It achieves competitive performance against commercial APIs and provides a full harness for whole-meeting transcription with cross-window speaker linking. The project is notable for its open-weights release, thorough benchmarking, and practical usability.
Target audience: machine learning engineers, researchers, and developers building meeting transcription or diarization systems
Repository: https://huggingface.co/Trelis/tiron
View on Hacker News