Is there a way to provide "prompt" to the binary?

[platomedical]

I have been using whisper-cpp so far. I am thinking of switching to faster-whisper-xxl.exe instead.

However, I don't see any option to provide a "prompt" via command line.

Also, is there a way to provide a path for the xxl_data folder? It is a very large folder - so I was hoping to store it somewhere else.

Thank you so much :)

[Purfview]

However, I don't see any option to provide a "prompt" via command line.

it's -prompt or --initial_prompt. I don't recommend touching it.
And there is --reprompt --hotwords ect.

Also, is there a way to provide a path for the xxl_data folder? It is a very large folder - so I was hoping to store it somewhere else.

Why you want to separate that folder? Copy everything together to location you want and add it to PATH, then you can call the program from anywhere.

[platomedical]

Thanks so much. My audio actually comes from another source. This program chunks the audio using its own VAD system - so I need to provide the initial_prompt containing the previous sentence.

Awesome work btw. I really want to make a donation (i don't need the Pro version - the XXL version is already perfect). But PayPal does not allow donations from my country. Any other way of donating?

[Purfview]

My audio actually comes from another source. This program chunks the audio using its own VAD system

Not sure what that means, but there is no need to use outside VAD.

PayPal does not allow donations from my country. Any other way of donating?

https://buymeacoffee.com/purfview
Or email me if you want Bitcoin address.

[platomedical]

My program allows the user to transcribe audio in a semi-realtime manner. It records the audio and chunks it using silero_vad in realtime. Each chunk is then sent to faster-whisper for transcription independently.

The initial_prompt helps - it allows me to send the text from the previous chunk to whisper.

Just made a donation via think link. Once again - awesome work!

• Vish

[Purfview]

Why not input whole audio and let internal chunking+VAD do its job, and you can use --batched to make transcription faster.

Just made a donation via think link. Once again - awesome work!

Thank you!

[platomedical]

That was my initial thought too. But I need to show the transcription progress while the user is still talking. So I can't wait till the end to get the whole audio and process it in a single shot.