How to Create Custom Audio Summaries of ANYTHING (That Sound Exactly Like You)

All right, welcome to another AI workflow tutorial. I always enjoy doing these, as this is as concrete and practical as it gets. Usually on this channel we talk about specific releases and tools that come out, but today we'll be combining a multitude of these tools to achieve something a bit a custom Notebook LM with your very own voice. I mean, sort of notebooklm has these podcast type conversations here. It's just one speaker, but it's fully custom. So you can change the prompts in the background, you can change the voices, you can change the style, you can change the workflow flow. It's set up to automatically fetch the files and create these summaries rather than NotebookLM, where you have to work manually with it. So it might not be exactly like NotebookLM, but as close as we could get. And it's fully customizable and almost free, depending on which components you pick here. Of course, if you want the highest quality, it'll cost a little bit. So essentially what happens here is you go from a meeting summary to a custom audio file that sounds exactly like you in both tone and style. And the voice. Instead of me just explaining this to you, let me show you the resulting audio file and then let's get into the step by step of how to set this up and. And do it yourself. All right, so this is the resulting audio file. It's a recap of a team meeting that we hold on a weekly basis. Anybody that misses it can listen to this instead of reading the transcript or a lengthy summary. Hey, team, welcome back. This is Igor with the recap of the AIA weekly team meeting from October 28, 2024. Okay, let's get into it. Raluka brought up the need for standardized PowerPoint templates for events with consistent branding. Matias supported the importance of color consistency, but noted that structures can vary. I confirmed that we have templates on Notion and suggested we collaborate on a new one. Okay, isn't this cool? I just want to note this voice is pulled from my news. You can use videos. So we extracted all the transcript and customized the style of GPT4.0 with a fine tune. We'll talk more about that soon. To sound just like me in the YouTube videos. Hey, team. Just to summarize topic four of our meeting, Matthias reported back on the success of our no Code Ethics talk and Ripple Agents events. So, as you can see, this is pretty damn good. And the very last thing before we get into this, I just want to say a big thank you to make for collaborating with us on this project. It's something that we were actually using within the company and when they reached out to collaborate, we were like, awesome. This is a great opportunity to show in great detail what's possible with their product. All right, with that being said, let's get into this and we'll begin with a little checklist that I prepared for you on what you will need to make this happen. First up, you will need a Dropbox account. That's how we built this automation. You can certainly customize it to work with Google Drive, OneDrive or any other file storage system you're using, but we wanted to switch it up. The last make tutorial on YouTube we did was with Notion. This one is going to happen inside of Dropbox. Next up, you need an account with make and here you can sign up for free with the link in the description below to give this a shot. You don't need to pay to actually try how this automation works and to set it up. If you're going to be using it regularly, you might want to upgrade to the $10 plan as I have done on this account here. Next up, you need like the blueprints that we prepared for this video, you can freely download this inside of our public Google Drive folder, also linked below, and simply import this entire automation into make, making it really simple to get up and running with this. Hey there editing Igor here. We tested the result of this video with somebody who had no experience in automation and it turns out it might be a bit too complex for somebody just getting into it. For that reason, we're actually adding a second file that will make it way simpler for you. So the original workflow is, hey, you need a Google Gemini meeting recording file, export it to Markdown, bring it into the Dropbox and then have the automation move it through the workflow and you get a resulting audio file. But if you don't have a Gemini meeting file, you would need to either custom prompt it in a way for it to change, or you would need to change the format of your document differently. That's a bit cumbersome and that's why we created the second automation, which includes the custom prompt that changes the formatting for you so you're not compromising any of the quality. Because to me that seems really important. That, hey, it starts with a formal intro, it starts with the date, it starts with a structure you want it to have so it sounds more professional. So you were not compromising all of that, but we gave you a second blueprints that will automatically change any summary you might have. Into this exact format that works with the automation. All right, I hope that makes sense. And let's get back into the more advanced tutorial. Just be aware. But there's a second file now in the description that is more beginner friendly. I would recommend you start that with that one if you don't have extensive experience with this stuff. All right, and there's two more items on my checklist. One of them is 11 labs account with a custom voice created in there. You don't have to create a custom voice. Hey team, welcome back. This is Igor with the recap of the AIA weekly team meeting. But it certainly adds to this project. We'll talk about that in a second. And then lastly you will need a fine tuned model inside of OpenAI's platform. In this case, we fine tune the GPT4.0 model on my own transcripts. Again, more details on that in a second here. Okay, so let's go through this checklist one by one and let's exactly show you what goes into these points and what it costs. So first up, Dropbox, I created a brand new account here. Just logged in with my Google email address and on the onboarding screen you can skip the paid plan. And if you do that you get signed up for free account with 2 gigabytes of storage that you can now use for this automation. Excellent. That's Dropbox right there. Secondly, same thing for make, you can just sign up with the link in the description below for a free account. Once you have your free account, as I mentioned, you will also want to get the blueprints that we prepared for you. Simply go to the link in the description and then download it to your local machine. We'll be using that in a second here. Okay, two more items left on the checklist. One of them is 11 Labs Custom Voice. Here you can sign up for free, but this won't allow you to create a custom voice. If you want to do the best version of this and you want your custom voice, I would recommend the creator plan that comes in at $22 a month. Although you can do it on a starter plan there. The voice just won't be as high quality. But essentially this allows you to upload a recording of your own voice, how you speak and then usually within 24 hours they will come back to you and then they will create the highest quality voice clone that is available on the market right now. I think right now this is the consensus opinion that eleven Labs is just the best at this. This is a thing that I won't be going through step by step here because it's really simple. Once you're on the paid planner, just head on over to voices say add new voice. And here under Professional voice clone you can upload a 30 minute audio file or longer of your voice and then they create it. I'll give you a tip though. One of the best things you can do is take some recording when you weren't purposefully talking just to record. So if you ever were on a podcast episode or inside of a meeting or you held some sort of lecture, that's a really good voice recording to take. Just make sure nobody else is speaking in there. You will want to manually edit out all the parts where other people are not speaking. Honestly, you need to do this manually. No automatic software is good enough to really catch everything and this file needs to be perfect. Then you upload it here and it creates a replica of your very own voice that you can then use in our make scenario here. Okay and last on our checklist and probably most excitingly, you need a fine tune. And fine tuning is something we talked about before on the channel, but it's something that most people overlook. And if you ever wondered how to get the very best custom style out of A LLM like GPT 4.0, well this is the way. And no matter how you prompt these models inside of ChatGPT or Fudo API, it doesn't matter what you put in there, it doesn't matter how detailed your instructions are, it doesn't matter how many examples of your writing or speaking you provide, a fine tune will always be the depth of the style replication that happens here. Now this is not a video just about fine tuning. But trust me on this, if you want to replicate your style, fine tuning is the way. And it is quite simple by now all you need to do is you need to go to platform.OpenAI.com, log in with your OpenAI account to create can be a completely free one and then you will want a few dollars in your account. $5 should suffice to fine tune a GPT4O model on your own writing. So just make sure under billing you add some credits here and then you want to head on over to dashboard fine tuning. And up here under create all you do is select the base model here and what you want to do is GPT 4.0. Not mini, not 3.5, 4.0. This works the best, especially for something a bit more in depth like the summaries that we're working with here. If you're on a budget or just want to play around with this, I would recommend mini as that is the cheapest. And then once you're here, all you need to do is give it its training data. And this is where really a lot of the magic with fine tuning is hidden. Everything else you can keep as default. But right now I'll put up a screenshot of the training data that we have used for my very own fine tuning here. This is a bit of an in depth workflow and inside of our community we're even developing an entire course on fine tuning because there is so many options here. But basically what we did here for this case is we pulled every transcript from the news you can use video that we have done for over a year now on this channel every single Friday. Then we compiled a big table of all those transcripts in different fields and paired them with input values. As you can see here, you can totally steal this formatting and adjust this to some of your own recordings. Again, you will need some transcripts of you speaking, so you can run some AI app to transcribe your next meetings. And from there you can pull out the way you speak, a transcript of you speaking, or you can put in your own writing tool if you have some. It's just important that it comes in as this format. And then what you can do, you can upload this Excel file to this free tool novelcrafter.com tools finetunes link in the description below. Also this is completely free and as you upload this it will turn your Excel file with all the examples into this JSON L format that I showed you which you can then upload into here. And as you upload this file and hit Create at the bottom, it will take a few minutes to train a brand new model like like these right here. And then you can link that model inside of the automation to use a custom version of GPT4O together with the custom voice from Eleven Labs pulling files from the Dropbox folder and all of that happening automatically. And that's a checklist. That's all you need to prepare for this. Once you prepare the items on the checklist, we can move on to bringing it all together inside of the Make Automation. And as we created this blueprint for you, you can simply create a new scenario like so inside of your account and and all you want to do is go to the bottom here, click the three dots and say Import Blueprint. I'll simply choose a file and here on my desktop I have the meaning Summary blueprint where I just click open and check this out. If I just hit save, it imports this entire automation for me. And now all the work here that is left is linking your various accounts that we prepared in the checklist step to this automation. And this thing can run in the background indefinitely as long as you have made credits in your account. So let's start with the Dropbox step as we need to do one more thing. For this to work we need a bit of a file structure here. And I'll do this by creating a few folders. The main one will be Igor's custom Meeting summarizer and inside of that I will create one that is called to do one that is called in progress and a third one that is called completed summaries. With this structure, literally all that is left now is linking all of it up inside of this automation. So I'll go to my first step, say create connection as I haven't linked the Dropbox account here and then as I'm logged in to Dropbox in my account already, I simply have to allow this and and the connection has been set up inside of here. I just have to navigate to the folder where the meeting summary files will be placed. Well, and that is the Igor's custom Meeting Summarizer folder that we just set up. And then inside of that we want a todo folder. If I ever place something in a todo folder, this step that watches files inside of that folder will automatically start this entire automation. Next step we pick our connection again. Continue. And then as you can see, a lot of these other steps will be preset by default. I just click OK here. This is fine. Again, I will not be going into every intricate detail of this automation and how to customize it. I'm just showing you how to set this blueprint and this automation flow for yourself. Okay, next Dropbox step again add the connection. Allow this and this. We have to do one thing which is set up the in progress folder as this moves the file from todo to in progress. So I'll go to Igor's custom Meeting summarizer and then here I'll pick the in progress folder. Excellent. This entire next row is something specific to Google Meet Summaries with their Gemini AI that we use. So all of these steps pretty much take apart the format that this meeting file comes in for our automation to work even more reliably. But if you're just coming in with a generic summary, you could just delete this entire row and directly link the Dropbox notes to this ChatGPT notes down here. I just left this here because this is exactly how we use it and how you get the highest quality results, which is what this is about. If you want to delete a part of this workflow, you can. If you want to change 11 labs to something that's free but lower quality, you can. If you want to change Dropbox to another file system that you might be using, you can. But this is how we use it, and this is the highest quality that we, we managed to achieve from it. So again, this takes apart the summary that comes directly from Google's Gemini Google Meets summarizer that you can see on the screen right now. It always comes in this predefined format. And this basically identifies the pattern and gives us the ability to work with this file in a bit more detail than just throwing a big paragraph at it and then chatgpt having to figure out what's going on in there. One pro tip would be you could totally prompt your LLM or your summarization step before this to export it into exactly the file format of this Google Gemini summary that I would also put into the Google Drive folder so you can see what that looks like. Now, look, I understand that your summaries might not be coming in this format, but this really works for us, and we found this to be the highest quality workflow. So what I'll also do, I'll give you a prompt inside of the Google Drive folder that you can use to summarize in a way that will produce this exact outcome every time. And those summaries will work extremely well with this. But again, it's not mandatory. You could just give it a paragraph of text. That's a summary. It's just an extra step to make this as high quality as possible, which is the goal of this video. I mean, NotebookLM is so good. That's why people love it. And we did what we could to get as close as possible here. As I mentioned, if you want to simplify things, you can always do that. Okay, so that is this row. And then in the next row, we will be going through the various ChatGPT steps. But for this, we'll have to set up a connection. I'll add a new one here. I'll just call this finetune model. And here I will paste the API key that I created here in my OpenAI platform by saying create new secret key. And then once you create it, you simply copy it over here to API key and say save. And then here what you want to customize is the model name Right. This is my own fine tune and you don't have my API key, so this won't work for you. You need to head on over to the playground and then take the model name up here. By the way, this is a different OpenAI account, so this won't exactly match up, but you want to put in that model name into this field. Everything else is set up. The prompt is set up. We connected it via our API key to our own account. But you need to link your custom model that you trained on your own data for this to work. So yeah, I've done that here. I'll click ok. You don't have to do anything in this step. In the next OpenAI step again, I just want to pick my fine tuned model connection again. Make sure the model is correctly filled in here. There's one more thing that you probably should customize in here, which is the prompt here, here. I gave it some context on me. I mean Igor Pagani, the founder of the AI Advantage community. So you probably want to change this first sentence to fit your own context. So ChatGPT recognizes you from the meeting summaries, right? I run these meetings and that's why I want the summarizer to Summarize this as me, Igor. It has my custom 11 Labs voice now. Also, ChatGPT knows what's up here. Cool. You can skip this tool step. In the OpenAI step again, we want to pick the fine tuned model. Make sure that the right model is in here again. And again, this is the same prompt. You might want to change this out to your very own context here. Just a quick side note on why multiple steps? Well, these different steps process different parts of the meeting summary and we just like this format of it. First saying hello to the team and then stating what the date is. Taking exactly the date that we extracted with this little step producing something that's really custom as opposed to just a generic meeting summary as you heard in the recording in the beginning, it greets the people and it states the date and then it goes into the points. Just these little touches that go a long way. If you're more advanced, you can go into customizing this. But if you just want to copy what I'm doing here, no need to change anything else in there. We can skip the tool step and then we're almost done here on the last row. 11 labs is quite simple. You just create a connection. I'll just call this My11Labs connection. And then all you want to do here is head on over to the 11 Labs account go to API keys, say create an API key, demo key one. I'll delete this one after this video. And here you can just copy this to the clipboard, paste it in here and voila. We connected our 11Labs account in which the custom voice that we trained will show up. Look at that professional Igor the AI advantage. This is the custom voice and how named it inside of the ElevenLabs account when training. And again everything else is set up. You can see it pulls in the transcription from the last step here already. And 11 Turbo V2 from our testing is the best voice that you want to use here. Ha. We're so close to being done and this actually working. So last thing I need to do is connect these last three Dropbox steps. So I'll just pick my connection here and then here I just have to pick a folder. I'll go to the meeting summarizer folder and I want to navigate to the completed summaries as we're at the end here. Now we want to generate this new Voice file from 11 labs inside of completed summaries. That's where the results go. I click OK and then the last two steps will be the same deal. I just pick my connection, navigate to my custom meeting summarizer folder and completed summaries, hit OK again. This saves the transcript in case you don't need that, you can delete the step and then again same thing. I need to pick my folder path, go to Igor's custom meeting summarizer completed summaries and then this last step moves the initial file that we sent from in progress to completed and that's it. I'm not saying this is super simple, but I believe that everybody can do this. It's just bringing together a lot of these tools that we talk about here on this channel. All right, so now what I could do is one of two things. I could set this up to run every 15 minutes. Ah, there's a little error here. It wants me to save this. I apparently didn't click OK here at the bottom. So now that I did that, I think we should be okay. So now all I have to do is click save here at the bottom. Now this scenario, this automation is saved inside of my MAKE account and I can set this up to run every 15 minutes like so. And now every 15 minutes it will check if there's a new file inside of the folder and run this automation. And as you can see now all I need to do is drag and drop this meeting summary file that I automatically get from the AI inside of Google Meets into this Dropbox folder, like so. And now inside of the scenario, I just run this once and we get the custom summary and my tone, style and my real voice. Well, elevenlabs, voice, all generated automatically. Here we go. Hello, everyone. Top G here. Welcome to the recap of the AIA weekly team meeting for October 28, 2024. Let's get started. Here's a quick summary of what happened at last week's event. Matthias shared his positive impressions of the presentation, noting the speaker's good vocabulary. Isn't that amazing? Now, you could imagine that there's a lot of customizability here. Change the voices, you could change the file system, you could change the prompting, you could do something different than summarizing meetings with this. I mean, really, this is an automated way to turn text documents into. Into audio files that sound just like you. And I think that's the coolest part about this automation. There's a lot of opportunities here. And on the MAKE platform, it's relatively simple and approachable to customize these things. I'm not saying it takes no effort at all. I'm just saying it's doable and as approachable as it gets. And that's why I like make. That's why I like teaching with make. So, again, a big thank you to MAKE for partnering with us on this video. I know this was a bit more advanced than we usually cover here on the channel, but I'm sure a few people are going to get a lot of value out of this. And just to round things out, I would like to say if you have any questions, probably the best place to ask them is the public area of the community. This is completely free. You can ask the questions here. Nevertheless, I would encourage everybody to give this a shot and to try this out with this video. You have everything you need to put this into practice yourself. All the links are in the description below. And with that being said, I hope you have a wonderful day and I'll see you soon.