0.9B Model Transcribes a 4-Person Meeting on a MacBook With 0.08% Speaker Confusion
Binary_orchid · reddit · 2026-10-22
A Reddit user benchmarked MOSS-Transcribe-Diarize 0.9B — a single-pass transcription + speaker labeling + timestamp model — on a 17.5-minute four-speaker AMI meeting, running on an M5 Pro MacBook Pro (64GB, mps backend).
- Performance: 714s of compute for 1,049s of audio (0.68x real time), fp32, greedy decoding, no VAD or hotwords, one model end to end.
- Diarization: All four speakers found; speaker confusion was just 0.08% at a 0.25s collar — lines are attributed to the right person almost every time; overall DER was 15.4%.
- Overlap handling: 87 of 317 segments start before the previous one ends, all different speakers — interruptions come out as overlapping lines with both people labeled.
- Caveats: One meeting only, scored on who/when; the author asks for pyannote 3.1 + Whisper DER and runtime numbers on the same files for comparison.
More from Infra
- Jensen Huang defends neocloud financing: high borrowing costs are a side effect of huge AI demand, not a business flaw — rohanpaul_ai · 2026-10-22
- PHP, Node.js, GitLab and LiteLLM hit critical security incidents the same day — alex_verem · 2026-10-22
- Bittensor IOTA cuts decentralized training merge time 5x to 6.7 min on 16B — bittingthembits · 2026-10-22
- Broadcom WiFi bug undisclosed since 2010 enables remote RCE on millions of access points — dyn___ · 2026-10-22
- Brain-inspired semiconductor chips aim for smarter, greener AI data centers — striketheviol · 2026-10-22
- Bittensor's Lium hosts over $30M in GPUs with 91% rented — bittingthembits · 2026-10-22