 So the New Yorker wrote this article about using whisper to transcribe audio into text and I thought it was pretty interesting. The author of this article said that it has pretty high fidelity and I thought I'd give it a try. I've been using whisper now for a couple of my videos to just to transcribe my voiceover into text and I've been very happy with the results. It's one of my favorite pieces of software that I use. So I wanted to show all of you how to use whisper for your transcribing. So step one is to go to the whisper I get a page and clone the project. My directory is code txt. This is where I want to put whisper. So whisper should be in here. I just cd into whisper and from here what you can do is just follow along with some of their examples. They say for a quick start you can just go ahead and use this command once you're in the whisper directory to download the base model. I've also tried using the medium model and I think it is a little bit better but it takes a lot longer. So if you're realizing that the base model is making a number of mistakes you prefer not to make then you can go ahead and download the medium model as well. But I'll go ahead and download this base model. This just pulls it from hugging face and now we have a base model. It shows us how we can use this command. I have a set of commands I really like to use when I'm transcribing. So this video is going to be a little bit of a spoiler. I'm going to show some of the transcription for a video I haven't released yet but that should be a little bit fun. So we'll give it a shot. Let's see if I can find my previous command in my history. Here we go. Okay so in this command I used the medium model. You can tell from this ggml medium. I'm just going to use the base and this hits a 16 hertz version of my wave voiceover. Most of the time when I do ffmpeg I'm not generating 16 hertz so I've got another ffmpeg command I'll use to get my audio into 16 hertz. Then I use a command for this is telling me the percent complete. I think this is coloring the output. This is saying how many threads to use. I'm saying that I want to output text and here is the output path. I think I got those commands correct. It's not all I'll look at the help page really quick before the video is over. Okay so it says there is no such thing here. I forgot to run the make command and they do show this in the documentation. They say that you can build with make right after you get the model. So we'll go ahead and just build with make. Okay the model is built and it conveniently output some of the commands so I can double check what I was saying here with pc print colors. I really like print colors. It shows you the confidence it has in the word that it picked. So maybe it'll be red if it's not so sure about that word and so when I'm reading through it I'll just make sure to pause on those red words and then pp is to print progress. So let's run this command and see what we get. All right so you can see that it wasn't completely sure about in. In this video I'm going to break down what all of statistics is about and it wasn't completely sure about everything in but I think it's doing a good job. Notice as I talk about chat gpt and it just keeps cutting away at the audio. One mistake that it made is this expert. What I really said is x bar. Maybe I just didn't enunciate but that was a mistake expert. It's supposed to be x bar. I did catch it down here so x bar was caught. Maybe I was just a little bit more clear in my speaking but this is a kind of transcription quality you get with the base model. Pretty dang good. It notices when I say things like x1, x2, x3. I'm impressed that it puts numbers there as opposed to words like x1. Could have been x and then once built out o and e. So that just saves me some time. Don't have to make those kinds of corrections. It also capitalizes proper nouns. That's nice. I don't have too much more to say about it. It's a nice way to transcribe and I do sometimes still go back and make some corrections but this is probably the best transcription software I've used and I'll probably continue to use it. So I hope you enjoyed this video and we'll see you next time.