zenithth Anibon Local Audio Transcription
Overview & Triggers
Use when YouTube has no subtitles or auto-captions for the target video.
1. Audio Extraction
Download audio stream as mono 16kHz WAV file (required for whisper.cpp miniaudio and to prevent >4GB WAV header overflow on long streams):
yt-dlp -x --audio-format wav --exec "ffmpeg -i {} -ar 16000 -ac 1 -c:a pcm_s16le audio_16k.wav && rm {}" "VIDEO_URL" -o "temp_audio.%(ext)s"
Or convert existing audio:
ffmpeg -i audio.wav -ar 16000 -ac 1 -c:a pcm_s16le audio_16k.wav
2. Local Transcription
Run GPU-accelerated whisper.cpp build:
Check for whisper-cli in system PATH or local $HOME/whisper.cpp build directory:
$HOME/whisper.cpp/build/bin/whisper-cli$HOME/whisper.cpp/main$HOME/whisper.cpp/whisper-cli
Model path defaults to $HOME/whisper.cpp/models/ggml-large-v3-turbo.bin.
Execution:
~/whisper.cpp/build/bin/whisper-cli -m ~/whisper.cpp/models/ggml-large-v3-turbo.bin -l th -f audio.wav --output-json -of whisper_output 2>&1
(-ot 540000 optional offset to skip silent start screens and avoid repetition loop bugs)
For full build options and platform configurations, see BUILD_WHISPERCPP_GUILD.md.
3. Format Conversion
Convert whisper-cli raw JSON output to pipeline-standard raw_transcript.json:
python3 ../cleaning-auto-transcripts/scripts/clean_transcript.py whisper_output.json --format whisper --output raw_transcript.json
Then proceed with the standard pipeline (chunking, signal detection, subagents).
4. Hallucination Detection & Recovery
Detect repetition loops / hallucinations via frequency analysis and auto-trigger recovery:
python3 scripts/detect_hallucinations.py whisper_output.json --audio audio_16k.wav -o recovered_transcript.json