AI Voice Conversion Tutorial : Sound like SRK, Arijit Singh, etc

I am going to use AI in order to make myself sound like Shahrukh Khan and by the end of the video you can do this as well first I'm going to record some audio that probably Shahrukh Khan might say nana I love you very very much but unfortunately I'm Don or Don kopakatnam a few moments later and in just a few seconds I am going to sound like this nana I love you very very much but unfortunately I'm Don or Don kopakatnam isn't that accurate the way Shahrukh Khan pronounces certain words the way his voice modulates when he's constantly talking before I teach you how to do this I want to show you one more thing let's take a song sung by one artist let's say pehli Nazar by Atif Aslam and let's make Arijit Singh sing it this is what the original vocals sound like [Music] and in a few seconds it's going to sound like this [Music] my name is sridev I am one of the founding members of 100x Engineers and in this video I will teach you how to convert one voice into another I am not a coder I'm not a developer I'm not even a software engineer but here I am teaching you how to use the RVC model because it's that simple and you will be able to do this in just five minutes so let's get started with the tutorial now I have a screen open in front of me which is basically a Google Document now RVC VC stands for retrieval based voice conversion RVC is a technique that uses deep neural networks in order to convert one voice into another voice for example I can speak something right into the microphone and if I have the data set of someone let's say Shahrukh Khan I can just essentially sound like Shahrukh Khan and this applies for any data set that you can create you can create a data set of even your own voice so I scoured the internet for multiple models and this is the best one that I have found it's done by this guy called Kayla maze so thank you kilometers for making this available for free and let's get right into it this document is essentially a guide that tells you how to set up etc etc you can go through this at your own pace but essentially I'm going to click this link which leads me to a Google collab page for those of you who don't know you can think of Google collab as your virtual GPU a lot of AI and machine learning projects require high computational power which is why gpus are recommended but gpus also come with a cost they're really expensive so Google made this platform form called Google collab which essentially acts as a GPU on the cloud where you can build very small scale AI models where you can train or run programs so this model is simple enough for it to be there on Google collab and we just have to follow a bunch of steps in order for you to be able to turn your voice into xyz's voice alright so Step One is essentially the installation so all you have to do is click the play button over here click on run anyway and wait for a few seconds now while our RVC model is getting installed let's take a peek behind the hood of what the RVC model consists of there are two forms of input number one is your voice input that you give through the microphone or an audio file that you can upload number two is the voice data set which is essentially who you want to sound like maybe you want to sound like Shahrukh Khan maybe you want to sound like Arijit Singh maybe you want to sound like Atif Aslam depends on who you like the way you can think of this dataset file is a collection of the person's vocals maybe this person gave a bunch of interviews maybe this person has sung a bunch of songs so these files are probably two to three are long recordings of just this one person singing so that the model can understand variations in their pitch and adjust its weights accordingly so we have two inputs number one is The Voice or the input file and number two is the data set and the data set layers over your voice and gives you the final output which is the modified vocals and that's essentially how an RVC model works which is the installation is completed as you can see now in the Second Step you are essentially booting the model you're essentially just kick starting the model in order to do this you need to give it a sample data set just to get it kick-started so now I'll go to this archive sheet which is essentially an Excel sheet that consists of thousands of models this Excel sheet alone has about 3 000 voice models of different people there can be cartoon characters that can be real life personalities as you can see there is SpongeBob over here there is Kanye West you can find some of your favorite singers there's Michael Jackson there's Freddie Mercury you can explore this on your own and this is provided by the same guy kailu means so let's take a random data set over here copy the link let's come back to the model and let's paste it and let's hit the play button [Music] all right model has been successfully imported that was very fast I'm telling you guys this is super simple I am not a coder and I am able to do this in seconds so definitely give it a try final step start the GUI and open the public URL all right we're gonna hit the last play button over here and a link will pop up in a few seconds which will lead us to our final graphical user interface all right we have the final link over here the link should look something like a bunch of numbers with the text gradio Dot live we're going to click on this voila you should see a graphical interface like this all right you have a UI over here not the prettiest looking UI but it will do because wait until you see the results and now let me teach you how to convert your voice into someone else's voice firstly you'll have to click on download model which is where you will have to import the data set so I've taken the PIN to go through the sheet and find the best data sets that I like the artists that I really like the people that I want to sound like and I've compiled them in this Google doc that you can see over here so let's go with Arijit Singh I'm going to click copy over here this is his data set which essentially is R's and R's of Arijit Singh just singing compiled into a zip file I'm gonna name this Arijit Singh and I'm going to click on download it'll take me a few seconds and here we have it now I go back to incidents I hit refresh and I should be able to find a rigid Sim over here you can select the person that you've just imported over here now step number two over here is to drop an audio file or to record some something from here for now I'm just going to drop an existing audio file I already have a bunch of songs that I have downloaded I've actually downloaded a vocals only version of pehli Nazar by Atif Aslam so we are going to start with an Atif Aslam song and make a rigid Singh sing the song so let's do that this is what the original song sounds like by the way foreign [Music] over here and we have selected Arijit Singh click on convert wait for a few seconds and there we have it are you guys ready to listen to this [Music] foreign [Music] I can't begin to say how freakishly accurate that is I'm gonna play it again just listen to it let's forward it a bit just listen to how freakishly accurate it is the way Arijit Singh sings the way he pronounces the words the kind of intonation that he has in his music the kind of modulation that he gives it's freakishly accurate foreign [Music] insane all that in just a few seconds now here's what I'm going to do I'm going to download this track and put it in the backing of the original song so that you can actually listen to pehli Nazar which is sung by Arijit Singh and tell me in the comments if you're able to find the difference foreign so that's essentially how you use the RVC model to convert the voice of a speaker into a completely different voice by layering it remember every time you want to load a new voice you have to go to download model you have to put the link of the data set over here download it click refresh and then you'll be able to see it on the list now I'm sure a lot of you will have this one question what if there is a voice data set that I want to use there is someone I want to sound like but the data set is not available in the Excel sheet that I just showed I have a few recommendations where you can actually go find data sets number one is hugging face number two is Google data sets chances are you will be able to find the data sets on these websites but if you are not if you are not able to find your favorite musicians or your favorite cartoon characters or your favorite celebrities data set you can actually train it on your own you can essentially download hours of footage of them speaking from YouTube or whatever and you can essentially make it into a format that you can actually input into this GUI how to do that how exactly can you train your own models how exactly can you train your own voice as well we are not going to cover it in this video but if you are keen on that let us know in the comments if this video gets 500 comments we will make another video on how to train your vocal data set even if it means that it's one of you copying and pasting the same comment 500 times that still qualifies all right that's it I hope you guys enjoyed this video I hope you guys will try the tool and come up with some really fascinating results if you create something exciting of your own do share it with us tag us on Instagram it's 100x Engineers we will repost it on our stories and we would love to see what you have created using the RVC model until then thank you and subscribe to 100x engineers

Channel: 100x Engineers

Views: 301,463

Keywords: Shahrukh Khan, Arijit Singh, Atif Aslam, Artificial Intelligence, Ai, RVC Voice Cloning, voice cloning, ai cover tutorial, tutorial, ai cover song tutorial, ai voice cloning, ai singer tutorial, fl studio hindi tutorial, cover ai tutorial, ai voice changer, voicify ai tutorial, ai music tutorial, voicify tutorial, voice clone, ai voice changer singing

Id: RJ0fgV-kXwI

Length: 10min 17sec (617 seconds)

Published: Tue Aug 01 2023