How to disable saving transcript results? (CLI)

[hustLer2k]

I want to just pipe transcription to another process without transcription results being saved.

[ryanheise]

You can't do that with the CLI (there is no option for it).

The easiest workaround would be to run whisper, then cat the file, piping that into another process, and removing the file afterwards.

A more involved way would be to write your own Python script, since then you could invoke result = transcribe("large-v2", "audio.mp3") method and then print it to stdout with print(json.dumps(result)) for example. That way no file will be written and the transcript will be printed to stdout. If you want to print out just the text, you can select that from the json before printing. Unfortunately you can't use any of the SRT/VTT/etc writers since those are all hard coded to write to files.

[hustLer2k]

Thanks. Is there a way to output the data in chunks from the script (to implement streaming in our system) instead waiting for the full operation to complete?

[ryanheise]

In Python, you could use transcribe("large-v2", "audio.mp3", "verbose=True") and it should print out the same progressive output to stdout that you see when using the CLI, but without saving the output to a file. However, since verbose=True is intended only as a progress indicator, there is no flexibility to control the format of this output and there is a chance that the format of this verbose output could change in future versions. If that's satisfactory, you could pin your app to a specific version of Whisper and check before upgrading whether it changes the format. If not, you could fork whisper and see where in transcribe.py it prints out that verbose output and you could modify that line of code to print it out in a different format.