Skip to main content

Audio ingestion

What this does​

If you have a radio broadcast / interview / podcast clip that matters for today’s assessment, this guide turns it into text evidence (a markdown transcript) and then runs the same analysis pipeline used for news and WhatsApp.

Transcribe audio → markdown​

npm run audio-to-md -- --input /path/to/recording.mp3 --date 2026-04-21 --station "KAN" --program "Morning show"

Expected: a markdown file is created/updated (default articles-audio.md) with the transcript.

If speaker labels are wrong, force “plain” transcription​

npm run audio-to-md -- --input /path/to/recording.mp3 --date 2026-04-21 --whisper

Expected: a single transcript without speaker diarization.

Analyze the transcript (add it to today’s assessment)​

npm run analyze-resilience -- --content-kind audio --files articles-audio.md --date 2026-04-21

Expected: signals are extracted and today’s report updates after refresh.

Troubleshooting​

  • Transcription fails with “file too large”
    • Cause: long recordings exceed the upload cap.
    • Fix: install ffmpeg so the script can split audio automatically.
  • Transcription fails (401)
    • Cause: missing/invalid OPENAI_API_KEY.
    • Fix: set the key and retry.
  • Analysis fails
    • Cause: missing ANTHROPIC_API_KEY (analysis) or upstream outage.
    • Fix: set the key and retry later if the provider is down.