Speaker 1: Whisper from OpenAI is undeniably their biggest contribution to open-source AI. Whisper gave software developers a state-of-the-art speech-to-text AI model that can power our own software projects. Well, what if we wanted to increase the speed of running these models without losing any performance or upgrading our computer hardware? Faster Whisper modified the Whisper models to run on CTranslate2, a high-performance inference engine for neural machine translation models. At a low level, this allows the same Whisper models to run in a format optimized for CPU and GPU performance, while requiring less memory resources compared to heavyweight libraries like PyTorch, which Whisper runs on by default. By using CTranslate2 instead of PyTorch to run Whisper, Faster Whisper transcribes audio to text four times faster running the largest Whisper model on a GPU. On a single virtual CPU server, I am able to transcribe the same audio files in less than half the time. This results in huge cloud compute savings for anyone using or looking to use Whisper within production apps. Inside of a Python file, you can import the Faster Whisper library. Same as OpenAI Whisper, you will load the model size of your choice in a variable that I will call model for this example. The difference here is we can specify the CPU or GPU to run the model and what compute type. By running in 8 compute, the model will run much faster than OpenAI's library forcing float 32 on CPU and float 16 on GPU. Here we see the major difference in the syntax is how we call the transcribed model. With Faster Whisper, our transcription will stream the segments of our transcription as the model runs versus OpenAI Whisper returning the full transcription upon completion. If you want live transcription, this is a huge benefit. For this test, we can use this inline for loop to store all the segments text inside of one string variable. Once the last segment has been added, I'll have the program print the text transcription to my terminal. Now, let's test how Faster Whisper performs in comparison. As promised by the developers, Faster Whisper performs more than two times as fast and performance increases improve further when running the large V2 Whisper models. To increase the speed AI Austin uploads 4x, deploy one click on this video's like button. If you are interested in a tutorial on how to code a full stack Faster Whisper AI SASS from scratch, click the notification bell to be updated when that video goes live. This has been Faster Whisper in 100 seconds with AI Austin. I will see you in the next one.