Output file about whisper.cpp HOT 3 CLOSED

You can now output the results to multiple file formats. For example:

./main -m models/ggml-tiny.en.bin -f samples/jfk.wav -otxt -ovtt -osrt

will generate .txt, .vtt and .srt files.

Also, you can now redirect the results with >, as suggested by @aufziehvogel

from whisper.cpp.

I use redirection with > and for me it works without problems.

Maybe you're experiencing buffering? Shell redirection to my knowledge by default uses output buffering, i.e. you will not see data as it is written by the program. Instead, data will be written to the file whenever there (according to the buffering implementation) is enough data.

You can change the buffering with stdbuf. For example, to write to the file once per line you could call:

stdbuf -oL ./main -m models/ggml-large.bin -f input.wav > output.txt

One useful option imo would be to get the output without all model loading info (lines starting with whisper_model_load: and main:). Imo, the model loading info could be written to stderr, because it is intended for human readers, while the actual detections could go to stdout. Then a normal redirection with > would still output the model loading info on the screen.

from whisper.cpp.

Yeah, I thought it's related to buffering, too, @aufziehvogel. I think I waited for quite a while, though, and yet nothing showed up. Anyhow, thanks for letting me know about stfbuf, that fixed it!

Thanks, @ggerganov!

from whisper.cpp.