Feature Discussion: Audio Transcription Support

[skogsbaeck]

Use Case

I'm building a therapy practice management system and needed to process
voice notes alongside documents. Audio transcription → RAG enables
searching across mixed media (PDFs + voice notes).

Implementation

I've implemented audio support following RAG-Anything's patterns:

• Added parse_audio() to Parser (like parse_pdf, parse_image)
• Optional dependencies (pip install raganything[audio])
• Config-driven (AUDIO_LANGUAGE, AUDIO_WHISPER_MODEL, etc.)
• Zero breaking changes

Questions

1. Is audio transcription in scope for RAG-Anything?
2. Would you accept a PR for this feature?
3. Any concerns about dependency weight (~500MB Whisper models)?

Code is ready if you're interested. Happy to adjust based on feedback.

[4ST-499]

Hi there! 👋

I would strongly support this feature request as well.

Use Case: I am currently experimenting with RAG workflows involving scientific lectures and podcasts. Having a native way to ingest audio files alongside text documents within RAG-Anything would be a massive improvement. It would streamline the workflow significantly by removing the need for external transcription tools.

Regarding the dependencies: I agree that keeping the core package lightweight is important. Your proposed solution of using an optional dependency (pip install raganything[audio]) seems like the perfect compromise. It gives users the functionality they need without bloating the installation for everyone else.

If the maintainers are open to it, I would love to see a PR for this feature!