Voicemail Transcription With Openai/whisper
Everybody,
I’ve recently discovered openai/whisper and have been trying in earnest to get this working with Asterisk for voicemail transcriptions
(Currently using the NerdVittles script with IBM Watson)
https://github.com/openai/whisper
After spending several hours today, I’ve successfully integrated my home Asterisk 16 voicemail with Whisper.
Once I have followed the instructions for setting up an API server
https://blog.deepgram.com/how-to-build-an-openai-whisper-api/
Initially, I setup a quad core VM to test this with, but discovered that without a dedicated card for the inference that it was horribly slow.
So, I’ve set up testing on my desktop (Kubuntu 20) since I have an nVidia GTX 1060 installed.
For the integration with Asterisk, I’m using a slightly modified script from nerdvittles IBM Watson script
sendmailibm
That can be found on their website
https://nerdvittles.com/free-asterisk-voicemail-transcription-with-ibms-stt-engine/
I will probably find a low cost nVidia video card and get a stand alone Linux box running to handle this project.
If you’re interested in the details, let me know.
Doug
2 thoughts on - Voicemail Transcription With Openai/whisper
Greg,
Just a note,
This would work if you have the API server running on a Linux x86 box.
Then Asterisk would be using curl and python to communicate with that API Linux box.
Doug
I really love this idea. Thanks for sharing. I’ve been looking for a good way to provide this service to my customers. Hopefully this will work for me too.
Thanks, Dave