omi icon indicating copy to clipboard operation
omi copied to clipboard

try nova-3 by deepgram

Open kodjima33 opened this issue 10 months ago • 13 comments

deepgram released nova-3 for multi-languages. let's try it

kodjima33 avatar Feb 18 '25 02:02 kodjima33

it's currently under the beta access. if you have the access of the beta then let me know. i would like to work on the implementation.

abhayguptas avatar Feb 23 '25 11:02 abhayguptas

@kodjima33 running backend shows up invalid imports issue:

File "/Users/abhishekkumargupta/Desktop/open_source/omi/backend/main.py", line 7, in from modal import Image, App, asgi_app, Secret ImportError: cannot import name 'Image' from 'modal' (unknown location)

Dont find these vars in modal module..

abhishek818 avatar Feb 23 '25 19:02 abhishek818

anyways, adding the option/param - "model='nova-3'" should be enough.

abhishek818 avatar Feb 23 '25 20:02 abhishek818

I want to try this

aliveevie avatar Mar 03 '25 21:03 aliveevie

According to the latest docs, Nova 3 multilingual support is still coming soon :/

related deepgram issues [1, 2]

Serdnad avatar Mar 11 '25 05:03 Serdnad

ok closing

kodjima33 avatar Mar 11 '25 06:03 kodjima33

@Serdnad could you find any replacements ?

beastoin avatar Mar 17 '25 08:03 beastoin

@beastoin need to try their new beta multi-language model - they've gave the access (check email)

kodjima33 avatar Mar 24 '25 23:03 kodjima33

on-it @kodjima33 / @thainguyensunya feel free to update the progress here a ;)

beastoin avatar Mar 24 '25 23:03 beastoin

Hi Folks,

The implementation to try Nova-3 multi languages has been deployed to Dev env. Currently Nova-3 multi languages support 10 languages: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch

If the input language from user is in the supported languages list, the STT will use Nova-3 model, otherwise it will use old Nova-2-general model

thainguyensunya avatar Mar 27 '25 07:03 thainguyensunya

notes:

--

sorry guys, the deepgram beta is not good enough to launch on omi production env - poor performance, ~1 concurrents connection on streaming.

close for now.

--

next: #1892

beastoin avatar Mar 31 '25 11:03 beastoin

hi a @thainguyensunya , please try the deepgram self-hosted

beastoin avatar Apr 04 '25 04:04 beastoin

Hi @beastoin

Please help to review the draft PR#2152. The change is also deployed to dev environment.

thainguyensunya avatar Apr 04 '25 15:04 thainguyensunya