Files

4.2 KiB

Troubleshooting (GETTR → ffmpeg → MLX Whisper)

Prerequisites not found

ffmpeg not found

Install on macOS:

brew install ffmpeg

mlx_whisper not found

Install with pip:

pip install mlx-whisper

Extraction errors

"This appears to be an image post, not a video"

The extraction script detected that the GETTR post contains an image, not a video. This skill only works with video posts.

"This appears to be a text/article post, not a video"

The GETTR post is text-only and has no media to transcribe.

"No og:video meta tag found"

Possible causes:

  • GETTR may be serving different HTML to different user agents, or the content is loaded dynamically.
  • Try opening the post in a browser and View Source; confirm an og:video meta tag exists.
  • If the page requires auth/JS rendering, you may need a different fetch method (e.g., browser automation) instead of a plain HTTP fetch.

Network errors / retries exhausted

The extraction script retries up to 3 times with exponential backoff. If it still fails:

  • Check your internet connection.
  • The GETTR server may be temporarily unavailable; try again later.
  • Try fetching the URL manually in a browser to confirm it's accessible.

Download errors

"Failed to extract audio" / "Output file is empty"

The source URL may not contain an audio track, or the stream format is unsupported. Try:

  • Verifying the URL plays in a browser or VLC.
  • If it's an HLS stream, ensure ffmpeg was built with HLS support.

ffmpeg download fails on HLS (.m3u8)

If the playlist uses redirects, test manually:

ffmpeg -hide_banner -loglevel warning -i "<m3u8>" -t 00:00:30 -vn -ac 1 -ar 16000 /tmp/test.wav

HTTP 412 Precondition Failed

If ffmpeg fails with HTTP error 412 Precondition Failed, the og:video URL has an expired or invalid signature.

Why this happens: GETTR's streaming CDN (stream.video.gettr.com) uses signed URLs that are dynamically generated per session via JavaScript. The Python extraction script fetches static HTML which may contain a stale/invalid URL, while the browser executes JavaScript that fetches a fresh signed URL and updates the og:video meta tag.

This is common for:

  • Streaming URLs (gettr.com/streaming/<slug>)
  • Live stream replays

Solution — Use browser extraction:

  1. Open the GETTR streaming URL in a browser and wait for the page to fully load (JavaScript must execute)
  2. Extract the og:video meta tag content from the rendered DOM:
    document.querySelector('meta[property="og:video"]').getAttribute('content')
    
  3. Use that fresh URL for the download step

The browser-extracted URL will have a valid signature and work with ffmpeg.

Manual alternative (ask the user):

  1. Open the GETTR page in their browser
  2. Use DevTools (F12) → Elements tab → search for og:video
  3. Copy the content attribute value
  4. Provide that URL directly to the download script

Private/gated GETTR posts (auth)

This skill does not handle GETTR authentication.

If extraction fails because the post is private/gated or requires JS:

  • Ask the user for a direct .m3u8 or MP4 URL, or
  • Use a browser/manual approach to retrieve the media URL.

Transcription issues

Hallucinations / repeated phrases

Use --condition_on_previous_text False (included in the pipeline by default) to prevent errors from propagating.

Poor transcription quality

  • Try a larger model: mlx-community/whisper-large-v3
  • Check audio clarity (background noise, multiple speakers)
  • Ensure the audio is in a supported language

word-timestamps not supported

Some MLX Whisper versions may not support --word-timestamps. The pipeline script automatically falls back if this flag fails.

HuggingFace 401 / Repository Not Found

If you see 401 Client Error or Repository Not Found:

  • The model may require HuggingFace authentication: huggingface-cli login
  • Check the model name is correct (e.g., mlx-community/whisper-large-v3-turbo)
  • Some models may have been renamed or removed

Quality tips

  • Audio-only WAV at 16kHz mono improves speed and stability (this is the default).
  • VTT output provides timestamps for building the timestamped outline.
  • Language is auto-detected; transcription stays in the original language.