main icon indicating copy to clipboard operation
main copied to clipboard

Can't chose default model for voice transcription

Open KirkPX opened this issue 1 year ago • 10 comments

There doesn't appear to be anywhere for me to chose what model I want to use for my voice transcription. I would prefer to run whisper locally, for example. Is this feature available or am I missing something?

KirkPX avatar Nov 06 '24 13:11 KirkPX

It's not available. At the moment, only OAI and Groq offer STT, and IntelliBar would use OAI if connected and fall back to Groq if OAI is not connected.

erusev avatar Nov 06 '24 13:11 erusev

Thanks for the quick response. Would be very cool if we could pass custom Instructions for STT to do some cleaning up at the margin.

KirkPX avatar Nov 06 '24 13:11 KirkPX

Would be very cool if we could pass custom Instructions for STT to do some cleaning up at the margin.

Do you have specific use cases in mind? Thanks!

erusev avatar Nov 06 '24 13:11 erusev

Yeah. Sample prompts.

  • When I say something, pause, and then repeat part of what I said, revise the output to clean up the repetition.
  • Where relevant, interpret the word "period" or "question mark" as me adding punctuation.
  • For longer run-on inputs, provide both the raw output and a more cogent and organized version that's casual, but still suitable for a business setting.

KirkPX avatar Nov 06 '24 13:11 KirkPX

Nice! I'll see what we can do.

erusev avatar Nov 06 '24 13:11 erusev

Yeah. Sample prompts.

  • When I say something, pause, and then repeat part of what I said, revise the output to clean up the repetition.
  • Where relevant, interpret the word "period" or "question mark" as me adding punctuation.
  • For longer run-on inputs, provide both the raw output and a more cogent and organized version that's casual, but still suitable for a business setting.

Do you know of an app that supports prompting transcriptions like this? Thanks.

erusev avatar Nov 07 '24 10:11 erusev

Yeah. Sample prompts.

  • When I say something, pause, and then repeat part of what I said, revise the output to clean up the repetition.
  • Where relevant, interpret the word "period" or "question mark" as me adding punctuation.
  • For longer run-on inputs, provide both the raw output and a more cogent and organized version that's casual, but still suitable for a business setting.

Do you know of an app that supports prompting transcriptions like this? Thanks.

There's a bunch out there that do a good job of handling context eleganty, but most have full access to your transcripts which I really don't like.

KirkPX avatar Feb 04 '25 18:02 KirkPX

It's not available. At the moment, only OAI and Groq offer STT, and IntelliBar would use OAI if connected and fall back to Groq if OAI is not connected.

So, I went in today and added Groq. I also went and deleted my key on openAI to try and coax the system to fall back to Groq. It looks like Intellibar is calling Groq when I do a transcription, but instead of returning the text, it's throwing an error to check the OpenAI key.

Is there a way to reset the OpenAI key? Perhaps I'm missing something.

Image

Image

KirkPX avatar Feb 04 '25 19:02 KirkPX

FYI for readers. A clean install fixes this.

KirkPX avatar Feb 04 '25 19:02 KirkPX

In the latest release, we don't include Groq as an option for transcription. We did this because it was complicating the implementation and might bring it back at some point.

astoilkov avatar Feb 19 '25 12:02 astoilkov