Speaker 1: Hello there, today I wanted to share another very useful, but potentially technical tool. Whisper. Whisper is a speech-to-text AI model made by OpenAI, the same people behind ChatGPT. It listens to spoken language and generates subtitles or a transcript for that audio. As a language learner, this is incredibly useful. There are tons of reasons you might want accurate subtitles for whatever you're immersing with. I talk about those reasons right here. But most things on the internet don't have subtitles, which is a bummer. Especially if you're learning a language where it's not super common, or a language where the official subtitles often don't match the spoken language. I'm talking to you, French. I find this tool most useful for movies and YouTube videos that I want to watch, but that I also want to have a transcript for. There are many other use cases I can think of, and actually just yesterday, somebody sent me a long voice message on WhatsApp, and I wanted to have it in text form, so that I could respond to everything they said without forgetting. It was really useful. Whisper isn't perfect, however. Not even close. In fact, their self-reported best language still has a 3% error rate. On this graph, the longer the blue line, the less accurate it is, the more errors. However, most of these inaccuracies are proper nouns, ambiguous words, or places where it kind of makes sense for it to be difficult. There are also sometimes issues with timing, but that's gotten a lot better recently. All this is to say, it's not perfect, but I certainly prefer it to nothing. When I was starting out with Czech, I really wanted to watch certain YouTube videos, but they were way too hard without subs, so even having mostly matching ones was really helpful for looking up keywords and getting more comfortable with the language in general. So let's look at how you can make subtitles for your learning. First, you'll actually need to have the file you want to process. I made a video here about how you can download YouTube videos very easily, but this could be done with a podcast file or movie file, whatever it is. In my example, I'm actually using an audio file from one of these videos that I recorded last week. Now, there's a few things we'll need. First, Python and then Pip. I've left some tutorials down in the description that you can follow. And we'll also need FFmpeg. That can easily be gotten with Homebrew if you're on Mac OS or on Windows, you might want to use Chocolaty. Those programs aren't strictly necessary, but you do need FFmpeg since that's essentially Whisper's ears, it can't hear otherwise. Tutorials for those are also down below. If you try things out and they're not working, join the Refall Discord server and politely ask for help. To make it easier for people to help you, say what you're trying to do, what you tried already, what kind of computer you're running, and what error messages and problems you're running into. And then people will be more than happy to help. Now, let's make sure FFmpeg is installed and up to date. Open your command prompt or terminal or whatever it's called, as an admin if you can, and run one of these commands depending on what type of system you're using. Hey, this is future Ben. As we were testing some of the installation, I realized that I didn't mention that you have to install Git. Thankfully, it's very easy. This is just how it knows where to find the file, essentially. If you're on Linux, you just need to run this command. If you're on Mac, you can use this link to download Git and then install it. And same on Windows with this command here. I'll make sure that this page is in the description. Once you get it installed, everything will run just fine like it does in the video. All right, back to me then. Once you have things ready, you can run the command pip install and then that line there. Or try pip3 if that doesn't work. And that will download the latest version of Whisper. If at any point you see an error like no module named setup tools underscore rust, you need to install setup tools underscore rust by running pip install setup tools underscore rust. If there's any module not found, just change the name. And with that, you're ready to make some transcripts. Let's take this audio from the other video. It's called audio.mp3. Very original. So I'll run this command on screen. This is telling it to run Whisper on this file, which can be found in documents, check, and then to use the model medium. This program needs to run on your computer, which means that it will make your computer work if you use too big of a model. The larger the model, the harder it works your computer, but the more accurate the transcription is. My computer doesn't really like the large model. So I just do medium. If you're having trouble getting it to work, try smaller and smaller modules until things work out. If you want to guess what's a good size for your computer, then check out this image. That VRAM column is what you can use to guess what'll work for your computer. Anyway, going back to the command I just ran, this is what it spit out. A bunch of different versions of the transcript that it generated. You can use these however you'd like. I'm usually after the subtitle and just generic transcript files, so I can delete the rest. But maybe you find a use for them. And before I go, I want to share a few tips for running this. First, it's best to run the command in the folder where you want the files to go. Because in that example, those files actually ended up in my home folder. And then I had to move them. But I could have done CD, Documents, Check Movies, and then just run the command. Next, Whisper appreciates it if you tell it what language the audio is in. It can figure it out, but it's not actually that smart and will just work better if you tell it. For that, add dash dash language, language. You can either use the two letter code or the English name for that language, capitalized. If you run whisper dash dash help, it'll give you a full list. If you know exactly what type of file you want, you can also tell it the output format, which is SRT or TXT or VTT or whatever. In general, transcript is TXT and subtitles are SRT. For expert users, there is a ton of stuff you can tweak, but that's all that's needed for us normal folk. So here's an example of an actual command that I would very likely use. Whisper, movie.np4, dash dash model medium, dash dash language check, dash dash output format SRT. And this will make subtitles for the movie in check using the medium sized model. That commands in the description of this video, so you can copy and paste it and then edit your parameters to fit whatever you're learning. But that's everything for me. If you're one of those advanced users, you might want to check out Whisper X as well, but I haven't had a chance to try it, so you're on your own, but apparently it's even more accurate. Anyway, thanks for watching this video. I hope you found it useful for adding subtitles to content. Have a great week and happy immersing. Hey, as you know, there's a lot of language learning advice out there, which can make it kind of overwhelming and difficult to actually get going on your own learning. If you feel like you're struggling to figure out language learning, you're not alone. It's an extremely complicated process with tons of different steps. If you're looking for a step-by-step guide to create the perfect language learning routine for you, then you have to check out our new course. We spent thousands of hours designing a simple and straightforward process that you can use to create your own personalized language routine that actually works. We understand that every learner is different and that you have to roll with the punches and adapt. Every day for 30 days, I walk you through everything you need to know to build an effective learning routine, no matter your circumstances. We give you the advice and resources you need to ensure your success, so you don't have to waste time looking for stuff to do and can focus on learning. And if any questions do come up, don't worry, we are always there to answer any questions and clear up confusion. And it's all backed by our 90 day, no questions asked, money back guarantee. If for whatever reason, something's not quite working for you, we insist you get every penny back. It's time for you to stop wishing that you could learn a second language. It's time to become the master of your language learning journey. Check out the link below to get instant access and start your journey today.