Phase 0 of the speaker-accuracy plan. sherpa fast-clustering threshold 0.5
over summed mic+loopback mono exploded a 2-speaker call into 83 clusters;
raise to 0.7 (larger threshold -> fewer clusters) and drop sub-700ms spans
whose embeddings are unstable and only churn labels in assign_by_overlap.
Mic and system audio are already summed into one mono stream before
diarization runs, so clustering alone can't tell which cluster is the
user's own voice. match_mic_speaker compares a mic-only sample's speaker
embedding (same sherpa-onnx model diarization already uses) against each
diarized cluster's own audio and returns a label->name map: best match ->
"You", the rest -> "Speaker 2", "Speaker 3", etc. Returns empty (no
guessing) when the mic sample is too short or no cluster clears the
similarity threshold.
iamdoubz