README: update --monitor emoji lists.

streaming: robust continuous stdin with full-state recovery

Fix WAV parsing for piped ffmpeg output

MPS speedups: various optimizations

Compact adapter buffer by reclaiming consumed tokens

Add --from-mic live microphone transcription and vox_stream_flush() API

Merge pull request #2 from 3podi/main

Document -I flag, latency/efficiency tradeoff, and updated benchmarks

Fix bandwidth spec: M3 Max 128GB is 400 GB/s, not 300 GB/s

SPEED.md broader credits rules, specify baseline hardware

keep last DEC_WINDOW in kv cache

remove for loop attn_mask

Pre-warm encoder + prefill MPS matmul ops at load time

Update SPEED.md with Q-tiled attention results (Attempt 14)

Q-tiled encoder attention: BQ=8 queries per threadgroup (encoder 23% faster for jfk)