VMware

This #raspberrypi is always listening… #openai

This #raspberrypi is always listening… #openai

#raspberrypi #listening.. #openai

“abe’s projects”

In this video I use OpenAI’s Whisper model on a Raspberry Pi to transcode audio from a conference room mic. It works pretty well, but it’s far too slow for a lot of applications. With some creative hacking you may be able to make it useful, I don’t know!

source

 

To see the full content, share this page by clicking one of the buttons below

Related Articles

22 Comments

  1. The hard part is trying to do realtime inferencing with Whisper. It takes 30-second chunks, so you need to detect mic input, read it, and then when the noise has fully stopped process all of it. This gets complicated because someone may be talking near it but not at it.

    I researched for weeks and couldn't find any method to have it just idly listening for a wake word. As far as I can tell there's no easy way for a consumer to do such a thing.

  2. You can use Cuda cores for this task, I tested it with a rtx 3050 mobile and the speech to text task is so close to real-time!!. In this moment I am doing some tests using a jetson nano

  3. Personally, I just use the API key for whisper. It's very very fast and super accurate

    Also in order to get any sort of good accuracy (better than Google's and iPhones) you must specify the language to English if you leave it on automatic, it's worse

Leave a Reply