Hi, Openai just release their latest whisper version. It's not better in ALL languages (heard mixed results about some asian languages but can't evaluate myself) but is faster and more compact. Switching the default to the new v3 version should be very simple. Potentially just switching the model name string and the documentation. Here's an example using faster-whisper-server
Cheers
Originally created by @thiswillbeyourgithub on GitHub (Oct 4, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/5906
Originally assigned to: @tjbck on GitHub.
Hi, Openai just release their latest whisper version. It's not better in ALL languages (heard mixed results about some asian languages but can't evaluate myself) but is faster and more compact. Switching the default to the new v3 version should be very simple. Potentially just switching the model name string and the documentation. [Here's an example using faster-whisper-server](https://github.com/fedirz/faster-whisper-server/issues/93)
Cheers
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
I made some tests here, the experience using the turbo model for STT even for portuguese, that is my main language, is very impressive. I can't wait to use it in openwebui
<!-- gh-comment-id:2393687509 -->
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
I made some tests here, the experience using the turbo model for STT even for portuguese, that is my main language, is very impressive. I can't wait to use it in openwebui
<!-- gh-comment-id:2394491854 -->
@thiswillbeyourgithub commented on GitHub (Oct 4, 2024):
> i will work on this. i think it should be user selectable.
I think it already is, sort of, you can change [environment variables](https://docs.openwebui.com/getting-started/env-configuration/)
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
yeah, I think that only is not possible to set the turbo version because it is using the faster-whisper to manage the model, possibly we only need to wait a new faster whisper version, but curently using the turbo model with transformers lib looks better than using the faster whisper
<!-- gh-comment-id:2394529675 -->
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
yeah, I think that only is not possible to set the turbo version because it is using the faster-whisper to manage the model, possibly we only need to wait a new faster whisper version, but curently using the turbo model with transformers lib looks better than using the faster whisper
@thiswillbeyourgithub commented on GitHub (Oct 4, 2024):
Hi, Openai just release their latest whisper version. It's not better in ALL languages (heard mixed results about some asian languages but can't evaluate myself) but is faster and more compact. Switching the default to the new v3 version should be very simple. Potentially just switching the model name string and the documentation. Here's an example using faster-whisper-server
Cheers
As I said there and unless I'm mistaken, it's already available on faster whisper
<!-- gh-comment-id:2394765445 -->
@thiswillbeyourgithub commented on GitHub (Oct 4, 2024):
> Hi, Openai just release their latest whisper version. It's not better in ALL languages (heard mixed results about some asian languages but can't evaluate myself) but is faster and more compact. Switching the default to the new v3 version should be very simple. Potentially just switching the model name string and the documentation. [Here's an example using faster-whisper-server](https://github.com/fedirz/faster-whisper-server/issues/93)
>
> Cheers
As I said there and unless I'm mistaken, it's already available on faster whisper
@allansouzasnowfox commented on GitHub (Oct 5, 2024):
Oh, of course! Sorry about that, I have the worst memory. Only just re-read the last message.
<!-- gh-comment-id:2395064990 -->
@allansouzasnowfox commented on GitHub (Oct 5, 2024):
Oh, of course! Sorry about that, I have the worst memory. Only just re-read the last message.
Default model will not be changed for the foreseeable future due to the model being much bigger than the current model, however I'll be modifying the audio UI admin settings to allow users to easily change the default whisper model.
<!-- gh-comment-id:2408731683 -->
@tjbck commented on GitHub (Oct 12, 2024):
Default model will not be changed for the foreseeable future due to the model being much bigger than the current model, however I'll be modifying the audio UI admin settings to allow users to easily change the default whisper model.
@tjbck wonderful feature! just looking forward to it. I test the feature, and wonder how to use gpu to run the whisper model inside open-webui.
<!-- gh-comment-id:2426950363 -->
@ice6 commented on GitHub (Oct 21, 2024):
@tjbck wonderful feature! just looking forward to it. I test the feature, and wonder how to use gpu to run the whisper model inside `open-webui`.
@thiswillbeyourgithub FYI, the variable name includes two underscores __ instead of a single one _. So WHISPER__MODEL instead of WHISPER_MODEL. Letting you know since the issue summary only has a single one.
<!-- gh-comment-id:2427037834 -->
@fedirz commented on GitHub (Oct 21, 2024):
@thiswillbeyourgithub FYI, the variable name includes two underscores `__` instead of a single one `_`. So `WHISPER__MODEL` instead of `WHISPER_MODEL`. Letting you know since the issue summary only has a single one.
@thiswillbeyourgithub commented on GitHub (Oct 21, 2024):
@thiswillbeyourgithub FYI, the variable name includes two underscores __ instead of a single one _. So WHISPER__MODEL instead of WHISPER_MODEL. Letting you know since the issue summary only has a single one.
<!-- gh-comment-id:2427086724 -->
@thiswillbeyourgithub commented on GitHub (Oct 21, 2024):
> @thiswillbeyourgithub FYI, the variable name includes two underscores `__` instead of a single one `_`. So `WHISPER__MODEL` instead of `WHISPER_MODEL`. Letting you know since the issue summary only has a single one.
Well your repo needs 2 underscores but open-webui only one, as per https://docs.openwebui.com/getting-started/env-configuration/
Or am I missing something?
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @thiswillbeyourgithub on GitHub (Oct 4, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/5906
Originally assigned to: @tjbck on GitHub.
Hi, Openai just release their latest whisper version. It's not better in ALL languages (heard mixed results about some asian languages but can't evaluate myself) but is faster and more compact. Switching the default to the new v3 version should be very simple. Potentially just switching the model name string and the documentation. Here's an example using faster-whisper-server
Cheers
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
I made some tests here, the experience using the turbo model for STT even for portuguese, that is my main language, is very impressive. I can't wait to use it in openwebui
@tcgumus commented on GitHub (Oct 4, 2024):
i will work on this. i think it should be user selectable.
@thiswillbeyourgithub commented on GitHub (Oct 4, 2024):
I think it already is, sort of, you can change environment variables
@allansouzasnowfox commented on GitHub (Oct 4, 2024):
yeah, I think that only is not possible to set the turbo version because it is using the faster-whisper to manage the model, possibly we only need to wait a new faster whisper version, but curently using the turbo model with transformers lib looks better than using the faster whisper
@thiswillbeyourgithub commented on GitHub (Oct 4, 2024):
As I said there and unless I'm mistaken, it's already available on faster whisper
@allansouzasnowfox commented on GitHub (Oct 5, 2024):
Oh, of course! Sorry about that, I have the worst memory. Only just re-read the last message.
@tjbck commented on GitHub (Oct 12, 2024):
Default model will not be changed for the foreseeable future due to the model being much bigger than the current model, however I'll be modifying the audio UI admin settings to allow users to easily change the default whisper model.
@tjbck commented on GitHub (Oct 21, 2024):
Added to dev!
@ice6 commented on GitHub (Oct 21, 2024):
@tjbck wonderful feature! just looking forward to it. I test the feature, and wonder how to use gpu to run the whisper model inside
open-webui.@fedirz commented on GitHub (Oct 21, 2024):
@thiswillbeyourgithub FYI, the variable name includes two underscores
__instead of a single one_. SoWHISPER__MODELinstead ofWHISPER_MODEL. Letting you know since the issue summary only has a single one.@thiswillbeyourgithub commented on GitHub (Oct 21, 2024):
Well your repo needs 2 underscores but open-webui only one, as per https://docs.openwebui.com/getting-started/env-configuration/
Or am I missing something?