S
Nov 10, 2026Speaker diarization on a MacBook Air M3: 4.4x faster on the GPU, 2.5x for the whole pipeline
Moving pyannote.audio diarization from CPU to Apple's GPU (MPS) made that stage 4.4x faster and the whole local transcription pipeline 2.5x faster. Numbers, code and what I did not measure.
Nov 10, 20266 min read0 reactions0 comments
