Speaker 1: It transcribes audio to text even faster thanks to the new feature that Subtitle Edit has added. Creatubers by double this week, and we have another video explaining this wonderful new feature, and we already have version 4 of Subtitle Edit, which in addition to adding some new options and making improvements to its interface, I love how it looks now, especially in dark mode, we have a new way to transcribe our videos in a much, much faster way. It is about Faster Whisper, an implementation of the C-Translate 2 engine, which uses all the power of Whisper in a much faster process than other engines. All the technical part and how it does it is explained in detail in its GitHub, with instructions to install it directly in your system if you have some skill. So let's get to work, because Faster Whisper is already available within Subtitle Edit, and this solito is the one in charge of downloading the necessary model and dependencies according to the components you have in your team. This makes the subtitling process via IA much easier. Let's quickly see what this new mode consists of. I have opened the mini panel video here, the last one I uploaded on YouTube, and if we go to Video and Audio to Text, Whisper, first we have to download FFmpeg, if we haven't done it yet, I have uninstalled and reinstalled Subtitle Edit so you can see it live. And when the FFmpeg installation process is finished, the next thing it will ask us for is the download of PerfViews Faster Whisper, which is the new engine we have available in Subtitle Edit. We click Yes, the engine will be downloaded, and now what the program will do is stay for a while thinking about analyzing the components we have in our team. In my case I have an NVIDIA RTX graphics card, so the program will tell me, hey, since you have a powerful graphics card, let's use the GPU to make this process much faster. So what it will ask me is to download two essential libraries so that it works well with my graphics card. I insist that this is a fully automated process and that it is not exactly the same in all teams. Depending on the components you have, you will have to download the corresponding libraries. After a few minutes we will have the libraries downloaded and it only remains with the Faster Whisper engine selected, select the language and then select the model. If you click on the three dots, a menu will appear in which you can directly download the available Whisper models. And now here comes another pain in the ass. If you remember other previous videos, I already told you that the larger the model, the longer it will take to process the audio to generate the subtitles. Well, Faster Whisper is so, so fast that even with the largest and most advanced model there is, which is the LARCH V2 that you see here, I'm going to download it, it is able to generate the subtitles at a devilish speed, so we will have much more precision in much less time. We download the model and what I do lately is to select to use the post-processing because as there are some corrections that I do that I don't like too much, I prefer to do them directly later. We click on Generate and for a video of 14 minutes and 21 seconds, it has taken only 2 minutes and 17 seconds. I insist, with the LARCH model, that is, the largest and most advanced, then what I do is press the combination CTRL F to correct common errors, I click Next and finally Apply corrections. I click OK and now you will see that some line jump corrections have been made, especially. And to top it off, I save the subtitles with CTRL S and that's it, I already have my .srt with the subtitles of the video applied with impressive precision in much less time, thanks to Faster Whisper. But is it really much faster than the previous modes? Let's see it. We are going to compare it with Const Me and we are also going to add a snippet that is DaVinci Resolve, which in the latest versions also includes a voice-to-text transcribing system. In the case of Faster Whisper, as we have already seen, it took 2 minutes and 17 seconds, and in the case of Const Me, the most advanced and similar to Faster Whisper that we have is the LARCH model, so we download it and in this case it took a total of 2 minutes and 50 seconds, 33 seconds more than in the case of Faster Whisper. And finally, in the case of DaVinci Resolve, we are going to select the Spanish language and here we cannot select the model, and in this case it has taken only 50 seconds to perform this same process. But the precision is less compared to Whisper's engines. For example, the word despotricar has not been taken well by DaVinci Resolve, while Whisper has done it perfectly, using both engines. Also, USB type C to USB 3.0 is perfectly caught by Whisper in both engines, not so in the case of DaVinci Resolve. And in the end, apart from the fact that in the case of Faster Whisper we have the greatest precision, a curious phenomenon occurs, and that is that in the case of Const Me, words are invented. This is something that has happened to me more times, because I believe that it takes similar models and expressions and Whisper expects it to say certain words and simply puts them in the generated subtitles. But I have not said this chau at any time. As for punctuation marks, I find that Whisper, especially in the case of Faster Whisper, does a more precise job, but that does not mean that DaVinci Resolve does not do a good job. But I like much more the flexibility that we have in Subtitle Edit using Whisper, because we can process by batches and make manual and automatic corrections very easily. And nothing else, direct and at the foot video to show you this new fantastic feature that Subtitle Edit has added and that leaves the audio transcription within the reach of almost anyone. And if you want to know how to transcribe audio using DaVinci Resolve, here is a video explaining it.