How to use directly the stdin option from yt-dlp stdout?

[davidbgk]

I'm giving it a try with youtube videos and I was hoping to be able to use directly the --stdin option like so:

uv run --with yt-dlp yt-dlp --quiet --no-warnings --extract-audio --audio-format wav --audio-quality 0 -o - "https://www.youtube.com/watch?v=7v6UbC5blJU" | ./voxtral -d voxtral-model --stdin

Unfortunately, it doesn't look to work without first downloading the .wav (without the -o - option of course) then parsing it (which actually works). Is there any restriction with how voxtral analyses the file or am I just doing it wrong? 😅

[dvhh]

I think the documentation is already providing the ffmpeg command line to transform your audio data in order to be accepted by the tool

uv run --with yt-dlp yt-dlp --quiet --no-warnings --extract-audio --audio-format wav --audio-quality 0 -o - "https://www.youtube.com/watch?v=7v6UbC5blJU" |\
ffmpeg -i -i pipe:0 -f s16le -ar 16000 -ac 1 - 2>/dev/null |\
./voxtral -d voxtral-model --stdin

[davidbgk]

@dvhh indeed, it does work piping through ffmpeg, thanks! It wasn't clear to me that it requires that extra step to deal with wav format as stdin. I don't know how pertinent it can be to update the documentation, at least there is that issue now :)

@antirez I keep it opened as a bookmark but feel free to close it if this is not pertinent to document.