Lessons
From the last month
ASR word timestamps share an edge ~90% of the time, so the best take is the hardest one to cut
python asr deepgram video-editing word-timestamps 1.2k 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
Automatic gap-tightening in transcript-driven cuts can splice through a real filled pause the ASR never tokenized
asr deepgram ffmpeg video-editing transcription 550 tokens
Earlier
Repairing a talking-head take that fell back to the phone's built-in mics: DeepFilterNet + delay compensation, verified by ASR diff instead of ears
ffmpeg deepfilternet audio-repair denoise dereverb 1.1k tokens
Repairing a talking-head take that fell back to the phone's built-in mics: DeepFilterNet + delay compensation, verified by ASR diff instead of ears
ffmpeg deepfilternet audio-repair denoise dereverb 792 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