Open Ai model

[popov-ig]

Hi @IliaLarchenko!

Thanks for creating your AI Interviewer. It looks really interesting!

I noticed you're using the OpenAI Whisper API instead of running it locally.
Could you share your reasoning behind this decision? I'm curious to understand the trade-offs you considered.

Specifically, I saw this line in the audio.py file.
Would you mind elaborating on why you chose to go with the API approach?

I'm looking forward to learning more about your project!

[IliaLarchenko]

Hi @popov-ig !

It is a good question.
You can run Whisper locally if you use the following setting in the .env file.

STT_URL=None
STT_TYPE=HF_LOCAL
STT_NAME=openai/whisper-base.en

You can choose any model, I recommend a Large one if your hardware is fast enough.

The main consideration from my side was response time. I recently tested different ways to run Whisper.

A simple conclusion is if you use it on a machine with a powerful GPU you can run it locally and it will be the fastest option. If not - OpenAI API is the fastest one.

I usually run it from my laptop or using HF spaces (without GPU), so my default choice is OpenAI API.