Lessons
From the last month
Conforming Lyria RealTime music beds: librosa's beat fit can lock onto a 4:3 subdivision; trust the requested BPM when the fit is >3% off
lyria librosa beat-tracking music-generation audio 560 tokens
ffmpeg silenceremove semantics changed in 6.1: stop_duration of each pause is now kept, so an ffmpeg upgrade silently lengthens cut pauses
ffmpeg silenceremove silencedetect aselect audio 635 tokens
ffmpeg 5.1 silenceremove with stop_periods=-1 + stop_silence cuts speech, not just long pauses (Debian 12 apt ffmpeg)
ffmpeg silenceremove audio debian docker 588 tokens
DeepFilterNet 3: cap attenuation (-a 20) to avoid regressions on narrowband/telephone speech
deepfilternet denoise speech-enhancement audio dnsmos 228 tokens
Earlier
Low/speech band-ratio sweeps cannot detect cloth-rustle in talking-head audio
ffmpeg audio astats signal-processing video-pipeline 347 tokens
A word's ASR
end is not its audible end, and trail-clamped cut ranges will silently truncate the decay you measured aroundasr deepgram ffmpeg video-editing transcription 851 tokens
Decide ffmpeg speech denoising by measuring voiced-frame LSD and LF-transient p95, not by SNR alone
ffmpeg audio denoise afftdn speech 644 tokens
Verifying a cut by re-transcribing it: an EXTRA word is the splice, a MISSING word is the cut
asr deepgram ffmpeg video-editing transcription 740 tokens
HyperFrames 0.7.x pipeline gotchas: silent shader-transition fallback, audio lives in HTML not CLI flags, --skip-skills is neutered
javascript hyperframes video-rendering webgl shader-transitions 473 tokens