Support transcription of videos (mp4, webm, etc.)

[fedirz]

https://platform.openai.com/docs/api-reference/audio/createTranscription#audio-createtranscription-file

[thiswillbeyourgithub]

Just stumbled upon this issue by chance. I just happen to have implemented code for this.

I made wdoc which is my tool for RAG (query and summary) that fit my precise needs. It includes various parsers, including some relevants to this issue:

• youtube links to playlists -> youtube links
• youtube links to videos (either using youtube transcripts, or translation, or downloading the audio then sending to whisper)
• url to a website where playwright is used to find a video -> download it -> whisper
• video -> audio -> whisper

Here's the link to the function to turn a video into text: https://github.com/thiswillbeyourgithub/wdoc/blob/d5a36ae02040420a698b361b7cb137a02d563af7/wdoc/utils/loaders.py#L1776

If you don't want to modify your code and just want to use wdoc directly you can do : wdoc parse --path my_video.mp3 --filetype local_video --only_text --audio_backend="whisper". Although now that I think about it I haven't set a way to overload the openai url for whisper. I'll add a env variable someday. Sooner if anyone's interested.