search

SevaSk / ecoute

Ecoute is a live transcription tool that provides real-time transcripts for both the user's microphone input (You) and the user's speakers output (Speaker) in a textbox. It also generates a suggested response using OpenAI's GPT-3.5 for the user to say based on the live transcription of the conversation.
https://github.com/SevaSk/ecoute
MIT License
5.85k stars 817 forks source link

Very Slow Load times, unlike the Demo Video #15

Closed aiking2222 closed 1 year ago

aiking2222 commented 1 year ago

Hi, While using the app the input and output loads very slow. its nothing like the demo video.

I have tried reducing the update time to 1 second still its not even close to being as fast the demo video.

Note - I am on free trial credits of the OpenAI API.

Am I missing something?

SevaSk commented 1 year ago

Is the transcription or the response slow? The transcription is done on your GPU unless your GPU is incompatible the it uses your CPU, which could be slow. Could you provide the console output?

aiking2222 commented 1 year ago

Thanks, I just read about how Whisper works. I thought it worked like gpt-turbo via API but now I understand it works locally to do the transcription.

My notebook has an Intel Iris XE which is an integrated graphics processor. Is that not compatible?

How can I make this work in my situation? Can you help. ( Please consider I am not a tech guy so your guidance, or pointing me in the right direction to make the transcription work in real time will be really helpful as I wont be able to figure it out by myself.)

SevaSk commented 1 year ago

The Intel Iris XE will not support Whisper GPU. Unfortunately, there is no simple solution to your issue. You could consider modifying the get_transcription method to use some API (somehow) instead of the loaded Whisper model.