Speaker 1: Hi, everyone. This is David. And in this video, I'll introduce you to Noscribe, which is an AI powered audio transcription tool that is powered by OpenAI's Whisper. Now, these are really awesome program with a couple of limitations. But let's go first of all, through kind of the write up. The creator is called Kai. And since he's in Switzerland, and the main purpose of this app is to transcribe interviews for qualitative research or journalistic use. So it uses Whisper and Whisper is this really awesome, super fast, kind of OpenAI Whisper model that is in C and C++. The main Whisper OpenAI uses Python, but it's good to have this. Whisper CPP is also in subtitle edit, and it makes transcription a little bit faster. Now, that's really awesome. They've also included pen notes to identify speakers, which is awesome, especially if you're doing interviews. This is really good for people who are transcribing interviews and need speaker detection involved. So it also includes a nice Microsoft Word interface to review, verify, correct the resulting transcript. Now, the main goal was to create an easy to use interface for my students and other non tech people, like my colleagues from the humanities department. So kind of tells us to spread the word. And that's what I'm doing right here. Now, at the moment, it's only running on Windows. Unfortunately, Mac and Linux, guys, sorry for that now. Obviously, these are the official page for NOScribe. And you can see it's cutting edge AI technology for automated audio transcription, a nice graphic user interface for OpenAI Whisper, and pen notes, speaker identification etc. So I'd like to talk about quickly just what you can see, it's free and open source runs locally on your computer, no cloud, no worries. Main purpose to transcribe interviews for qualitative research or journalistic use. It can distinguish different speakers and understands 99 languages more or less. So it's basically what OpenAI does, includes a nice MS Word interface to review, verify and correct the resulting transcript stands on the shoulders of Whisper. From OpenAI, Whisper CPP, Badgeology and Peernote. And then this is what it looks like. And then their limitations. Probably number one, it's only on Windows, Mac support is planned and somewhere along the way, Linux. Since it uses sophisticated AI models, the download is quite large, about 3.7 GB, because they're using the large model to operate this. Now, just to throw this in before you go and look at the app, I wish they gave us the option to self download the different models because it would be easier to actually download the smaller models, probably the based medium from the Whisper CPP, this particular, because I believe there are a couple of a couple of these particular models that have been prepared by Geology and yeah, some some of these bases, I believe they are in I believe it's Hugging Face. Yeah, Hugging Face. You can kind of find the different models if you'd like to download. So you can imagine downloading the large model, almost 4 GB, 4 GB here. That's why we're getting this that is about 3.7 GB. And yes, it is it needs a decent computer with a smaller models, you don't really need this. So it would be awesome for Kai to kind of give us the option to choose what model we want to use, or even self download, instead of having the app being like 3 GB, consider letting it run overnight on a slower machine. So that's, that's what I'm saying, poor quality audio will lead to poor transcription results. That's obvious with any AI. No automatic transcription is perfect, there'll always be some manual revision, obviously. And there are a couple of factors that affect why they named NOScribe about Kai. He has a PhD in sociology with a background in computer science, qualitative researcher and teacher at different universities in Switzerland and Germany, installation. So I'll just take you over to the switch drive here for the download, because it's really huge. And the reason why it's huge is because of that particular, for example, this is a version I downloaded, I believe, is about 3.6 GB, really, really huge. It would be awesome if we can actually have the option to auto download the different kind of models right inside the interface. So I believe I'm almost done with kind of going over this. Yeah. So all these are, I'll leave them below for you to read. Now let's look at it. Here it is. It looks really good. As I've mentioned, I wish there was an option to kind of choose the model I want to use, because it makes things much easier. Secondly, in terms of the transcript, it only saves in MS Word, I wish it's possible to kind of save as a text file, that would be awesome. Also, as much as this is for transcribing interviews, I wish it was possible to kind of auto generate subtitles using this. So the options for creating subtitles would be awesome. So let's begin by selecting a file, I'm going to actually test this file. And I'll just copy the name. Because of the next step, click on Open, it's going to be loaded right inside here. And then you need to select a file name. So I wish it's it picks the name of the video that you set up right here. Let me just paste it right there and click on Save. So there we are. So I've got the audio selected, it's in this video, the transcript file name, everything is set. And you can see here some few instructions that if you want to start and stop at like the third minute, it's going to transcribe the first three minutes. So if you actually just want a selection, you can make your settings right here. Now the language, you can either set it to auto or choose from the languages available here. And then speaker detection, auto, that's okay, or none, quality, precise or fast. So you can choose that. Let's go for precise. One thing I'm 100% sure is that it's going to take about three to five minutes, but I'm assuming it's about five minutes. Let's click on Start. It's converting audio, audio conversion finished. Speaker identification, it's loading Pinote. It's going through some segmentation, speaker counting 100% I believe it only knows it's only me. And then we have kind of like now the next step that is going to load, they're calling it embeddings. Not sure why it's being called that, but it's more or less transcribing. Now, as we've mentioned, this is heavy, because I believe it's using the large model. And that large model is a little bit heavy on the system. That's why you're seeing the not responding kind of icon up there. But in the background, it's operating as it should. My fans kicked in a while back, but it's loading. So that's why I'm mentioning that if it's possible to kind of have the option to choose whatever model one wants to use, the models from Georgy are really well done. And they're super fast, depending on which one you use. The base and the medium I believe should be faster than the large model. The large model is much more accurate, if that if accuracy is something that you're going for. But point to take note is that the large model does take a longer time, even when you have a beefier machine in terms of specs, and all that, what we're going to do is we're going to give it a couple of seconds, kind of minutes for this and wait and see the results that we get. So let's just wait for it to kind of keep going. And as you can see, we just keep getting not responding as it moves from one kind of percentages down here to the other. So 47 moves to another percentage, and then the not responding goes 52. So for a one minute, like 42 second file, if I was doing this in subtitle edit, it's going to take me with just the base Whisper CPP, it's going to take me a very short time. That's why I'm saying that I wish we can have something much much lighter in terms of the model being used. It also reduces the size of Noscribe as an application and just gives people the kind of notification that hey, if you'd like to use X number, or this particular model, then you need to download it and use it. But I feel that I like the interface. And I feel there's a lot of potential for this particular application Noscribe in terms of audio transcription for interviews, even subtitling, automatic subtitling, I believe it is something that can be really really useful, especially for people who do interviews, research and such kind of stuff, or even generally people who do audio transcription and subtitling. So let's give it a couple of more seconds to see where we are at. The bar right here, the sigma right there is at 94 for embeddings. Let's see what else we get. It's loading now it's beginning the transcription, Whisper CPP. So this was speaker identification, it's going it was going through the whole one minute, second clip. And that's why I'm saying that probably, if I turn this off, it would have been a little bit faster. But I feel that using smaller models of Whisper CPP is going to make it much faster and less heavy on the system. So my fans are in full blown capacity here. And let's see, let's see what the transcription gives us. Something else I've noticed is that once it starts doing the transcription, it's giving it in kind of like a breakdown of what is going on. It's like transcribing a section and then another section and another section. That's what it's doing. But maybe there's some there's some room for improvement, maybe a kind of an interface where the video can be seen maybe somewhere here, instead of having all these write ups here, knowing what is going on here, maybe some kind of things that can be seen. So this is what it looks like. In this video, I'll show you how to take high quality screenshots or image stills in DaVinci Resolve. Welcome to this video. My name is David and I hope you're well wherever you are around the world. Now, I'm in the edit page. And one of the few things that you need so that you can take high quality screenshots is ensure that when you go to playback, timeline proxy resolution, that it is set to full. If it's half or quarter, then you will not get the full resolution of that particular obviously, I believe it's screenshot. But based on what I've read of that particular transcript, to where it is, that's really really perfect. English is not a fast language to me, but it's perfect. And the reason why I'm kind of recommending it although it's taking long, is that it is perfect. Let's continue particular image. So it's set to full. Next up, we need to select the color page. So I'll click on the color page here, just right here. And we go into the color page. Now when I hover over this particular image, and you can actually even scrub through to probably look for a new screenshot section, then hover over this, just right click and then click on grab still, click on that. And once I do, that particular transcript is really, really well done. It's taking some time, but I feel that the quality is really, really good for this particular. So and once I do that, we've got everything here. And all we need to do now is just right click on it and click on export, click on that. And then we can call it whatever, for example, screenshot, then click on export. And that's all. Now let's see our screenshot. This particular transcript is really well done. And as I was mentioning, it would be awesome to have like an interface to actually play the video even if you wanted to create subtitles, or even if Notescribe morphed into a transcription and subtitling application would be really awesome to have that. So the only issues I've mentioned, it's taking a lot of time for one minute 40 second clip. Even if I was using subtitle edit, this is done in like a minute. So it's done. Let's just double click on it. That's high quality screenshot from DaVinci Resolve. Super simple. And as you can see, it's super high quality and it's full resolution, you can see almost all the details that you need to see. And that is how you can take screenshots in DaVinci Resolve. I hope this video is of value to you. Thanks for watching. So the biggest point here is that Notescribe is a very powerful AI powered audio transcription tool, powered by OpenAI's Whisper, Whisper CPP by Geology, and also PerNote for identifying speakers. So for me, it's only one speaker, it's just S00. The only unfortunate thing is that it's taking a long time for a one minute 40 second clip to take six minutes. That's really long, but the transcript is almost perfect. If I'm sure there's something perfect to me, so I don't think I would add anything else other than a couple of options, increase or save the transcript as a text document or document, have the option to kind of save subtitles, be able to set the number of lines one wants for the subtitles, the number of like characters per line, depending on what you want. And then having the option to kind of auto download the models right here instead of downloading a 3GB file to be able to download something like a small file like the Notescribe itself. And then if you'd like to download the models later on, then you can do that. So that's Notescribe. It's really awesome. Try it out has its own limitations, only available on Windows, huge download for that matter. And I'm sorry, I think it's either it's my internet or maybe where the file is hosted. But this file took a long time to download a really long time. And I do believe I have something, a good internet connection. So that's it for this video. Until next time, stay safe and never stop learning. Thanks for watching.