How to Look for Plagiarism in (AI) Music - Complete Guide

Hello and welcome back to the channel, so I was playing around with the beta version of Yudo on yo.com and after a few iterations of trying to generate a standalone FK song, the

music

generator created a tune that I really enjoyed listening to and that I thought which might have some potential as a

music

production or music publishing in its own right, so I thought why not also make a music video using the AI ​​tools for image and video generation that are currently available to us and see what I can come up with and that's exactly what I did.

Basically, I put together a 100% AI-generated music video. I now plan to continue developing this project and will post a separate video about the processes and workflows that got me to where I am and also. The steps I intend to follow in the future, but in the meantime, you can watch the full music video and listen to the song called Dances of Daylight by skipping to the last chapter or with Tim's stamp in the description below. uh, I might do that too at some point. publish the finished product as a standalone music video and in that case I will make the link to that video available in the description as well according to udo terms of service.

They grant all rights to the music that the generator produces, but also very clear about the fact that they will not be responsible for any copyright claims that the music that is being generated may incur and to be honest, I think that's fair enough, just as traditional musicians can even unintentionally fall into the undesirable situation of writing a melody. that may be too similar to some other song that has already been published and therefore runs the risk of being subject to copyright or intellectual property. intellectual property claims. AI generated music would probably be equally prone to

plagiarism

, although I believe that since the developers of the AI ​​generators have access to such huge data sets of original music that the generators would be trained and able to avoid copying the work of someone else, but these are just assumptions, so I will try to check if the piece of music I achieve through udio may have chances of running into intellectual property issues with existing copyrighted material.

To be honest, this entire tutorial can be used by anyone who wants to check if their music is plagiarizing someone else's music, as these strategies would apply not only to AI generated music, but also to AI generated music. Certainly, they can also be used to verify music composed with more traditional methods. Through my research, I found three different types of

plagiarism

detectors and we will use plagiarism detector as a general umbrella term, although in reality these tools and strategies are I will see that they all serve different purposes and produce different results overall, although it is Safe to say that these detectors specialize in one, detecting plagiarism and the musical composition of the song, two in detecting plagiarism, plagiarism and the lyrics of the song, or three.

look

ing for plagiarism by exclusion with song recognition applications now I only found one tool that was actually able to perform the task described in point 1 which is basically decomposing and analyzing the musical composition or rather the vocal and instrumental components of the song and then compare Unfortunately, this tool is in beta testing and while it seems to have a lot of potential and is definitely worth keeping an eye on, I think it is still very mature and not ready to deliver reliable results.

When it comes to effectively detecting plagiarism in music, the website I'm referring to is called MIP or MIP on ai.com, I'm not quite sure what the acronym means. It appears to be a new website that just started in January. Ary from this year and appears to be based in Korea, which, as you will see later, may be a relevant piece of information. This online application basically allows users to submit songs they want to check out and in return, the software makes reports on how. They closely match the songs they are reviewing to existing published works, the way their software works, as their website says, AI divides thousands of songs into four bars and compares their substantial similarities to good songs with high plagiarism rates and then continues saying through AI music analysis technology, it analyzes the song structure, rhythm, melody, harmony of music to detect plagiarized songs, among many other songs.

I'll quickly

guide

you through the process of checking a song for plagiarism with this online app so you can see firsthand what it

look

s like. As the whole operation is quite easy, I should say that you can still sign up and use the software for free at the time of recording this video and I imagine that is because it is still in beta testing. All you have to do once you're in is decide if you want to check vocals only, instruments only, or both and simply upload an MP3 file or provide a link to the song you want to check for plagiarism.

Once the app has finished analyzing and comparing your track, you'll get a full report on how similar your song is to any existing tracks that you imagine would be included in its database of original copyrighted works or in the group. of songs they may be pulling from the Internet, as each song that will be fed into this machine will likely have some degree of similarity to other existing releases. At the bottom of the results page, they also provide a baseline scale for how likely the song would be to run into copyright claims. If you want to analyze the data further, you can click the check button below. to each song which will upload a PDF file where the entire song is divided into four bar segments and also shows which segments have the highest level of similarity to the reference track and would therefore be potential culprits of plagiarism, as you can see The results of My Song Dances of Daylight generated by AI would be quite reassuring since the most similar song according to the app, Blow Me One Last Kiss by Pink, has a plagiarism rate of around 22%, which is in the safe zone according to their plagiarism rate scale that I have.

To say that the AI ​​generated song I'm reviewing doesn't sound anything like the pink song. I can't really do a live comparison here because I don't want to risk a YouTube copyright strike on this video, but you'll have access to both. my song and the other songs that were cited as possible copyright infringement so you can judge for yourself just as a point of reference. I also checked out Winter Winds, a song by Mford and Sun, uh, which is pretty much the same Indie Fulk genre as Dances of Daylight, my AI. generated song and also roses Blow Me One Last Kiss, which was the song with the highest plagiarism rate, compared to my AI generated song.

One thing that was a little alarming for the Mumford and Sun song is that the app didn't actually find m and Sun's uh Winter Winds in its reference database which seems to come primarily from SoundCloud and YouTube. I should point out that the Mumford and Sun song is quite popular. I would say so far it has collected around 23 million views on YouTube and more. 96 million plays on Spotify, so it's a little worrying that it wasn't detected during this check, on the other hand, although the app found a lot of K-pop songs if we look at the results for Winter Winds. we can see that it performed pretty well overall, the song it was most similar to had a 26% plagiarism rate which is still in that safe zone according to the website's plagiarism scale as a benchmark, although I must say that still performed worse than my AI-generated song got a 22% rose, Blow Me One Last Kiss on the other hand did absolutely horrible, apart from getting a 100% against itself, which obviously won't be taken into account. count as a result, the song ranged between 37 and 54%. uh for the tracks that it was compared to, I think we can get an idea of ​​what's going on and where the website may be going wrong uh right here in the fine print um where, right under the plagiarism scale, it says that there may be differences depending on the genre to me, uh, and this is just a guess which means that the results will vary depending on the genre of the song being checked because most of the music they use to test songs is limited to a bunch of particular genres, probably mainly hip-hop, EDM and definitely a lot of pop, this website works very well technically as a web app and the idea is top notch, but I wouldn't trust the results to be producing, at least not yet , once again, seems quite mature. in my opinion, and there are some warning lights that can't be ignored, like not being able to listen to a fairly popular song with millions of views and streams like mum and Sun's Winter Winds, it all leads me to think that unfortunately it's not all that.

It is reliable, however, I have to reiterate the fact that it is a brand new website that is still in beta testing. I definitely think this tool is worth keeping an eye on, though for future improvements, now the second type of detectors I found for letters. They were all text-based and, although not all are designed specifically for this purpose, they can be used to check song lyrics for signs of plagiarism. I tried a few of these, but the two that stood out are the Grammarly plagiarism checker on Grammarly. .com plaguerism D Checker and qex Checker cex.com are free and neither requires you to register grammatically.

You will also take a test to find writing problems while looking for plagiarism and, to your advantage, the interface. it is very simple and easy to use plagiarizing in my case are actually the Bible, so I don't think I have much to worry about. To be honest, I think most likely all of these tools use the same underlying API, so they're probably pretty good. They are very interchangeable, as you can see, the lyrics to my song scored pretty low on all the plagiarism tools I used, which means I don't think I have to worry too much about the lyrics being subject to copyright, finally We can move on to the third and final type of plagiarism detection tool that I mentioned above, they are known as song recognition or song identification applications and they would be used to determine the risk of plagiarism by exclusion, so basically, if the application of recognition does not identify your song, it would be an indication of the fact that it is not included among the millions of tracks contained in the application database and therefore one could conclude that it is very unlikely that the song has been copied.

There are a few song recognition apps worth mentioning, the most popular would probably be uh. Be Shazam, you can download it as an app on your smartphone, but you can also add it to your Chrome browser as an extension. Shazam is also built into iOS devices and can be accessed through Siri or its control centers. The caveat with Shazam is that the audio fingerprinting method it uses to detect music matches is only really possible if the song recordings being compared are exactly the same, so there really isn't much wiggle room. For nuances, a big contender in this space that we should definitely keep an eye on, Google appears to be investing significant resources in developing its own AI music recognizer.

Currently, you can access its music recognition features on Android devices through Google Assistant or the Google Search widget. Google developers seem to be taking a slightly different approach compared to Shazam allowing for greater versatility and recognizing musical compositions and melodies to the point where you can simply hum a melody for the app to detect and return a list of possible matches that It also shows a percentage of the probability that the match could be for each one. From the results in our case, it is understandable that this type of flexibility would help increase the probability of finding possible cases of plagiarism.

I also found two web-based apps, the first one is quite easy to use and is called aha music, which is aha music at aha. music.com and really all you have to do is play some music for the app to detect it or you can even upload an audio file and it will analyze it and try to find possible matches. The second web application appears to be alittle less. accessible than the first, but also seems to be a bit more advanced, it seems to be used as an API for application developers, although it can also be used as a standalone web-based song recognizer.

The website is acrcloud atacr cloud. .com, as you can see, they have a database of over 100 million tracks, accessing the app is not that simple, although you would have to click the start trying now button to sign up for a 14-year free trial. days he won. You will not be asked for any credit card details, but you will be asked to provide some personal details including the country where you live and your phone number, then you will be directed to this all products page where you can select the scan option of files at the bottom right as you can see it says scan and generate scan reports for uploaded files.

You will then be directed to the containers page. Here we will have to create a container by clicking on the blue create button at the top left and then on the dialog box that appears. give the container a name, choose the audio engine. I'm going to select the audio fingerprint and cover song ID, then we can choose the audio source. I'm going to choose a line in the audio that is the audio from the original file or the noise-free stream, but, uh, you can also try the recorded audio option. By default, there is only the ACR Cloud bucket that I think should contain the original ACR Cloud song database, so we'll choose that one.

For the scanning policy, I would choose Spot Scan since I will be sending files containing a single song per file, I would leave the callback URL blank and select enable music voice recognition and then press the confirm button once the container. You can click on it and from this page we can upload an audio file. or a fingerprint and I'll touch that in a second or you can link the music online once you submit your request you can check the status you will probably be processing, to update the status press the update button on the right. from this top filter bar here, once the status is ready, you can select the tract and download the report by clicking on the actions tab and then selecting export from the dropdown menu.

The report is basically an Excel spreadsheet that contains all the details related to the song. that has been acknowledged, including the song title and authors and links to major streaming platforms such as Spotify and YouTube. If the status is no result, it means that there were no matches for your search and the report you will get. obviously it will be blank the price of this app is not high at all, they use a pay as you go model with no minimum fees and the cost is basically a fraction of a penny or a penny for each valid result you get to further evaluate your clues of music that use music recognition, you may want to go as far as dividing your compositions into different musical themes, at least one for the vocals and one for the instruments and submit them to the identification tool separately or you can also try cut them. the song into different parts or segments which could be, for example, the verse, the bridge, the pre-chorus, the chorus and test each segment separately to see if you can detect similar songs compared to the particular section of that track what you are sending.

Acknowledgment comes with some limitations, particularly in the context of its use to check for plagiarism. The song being detected by the recognition app would normally have to have the same so-called audio fingerprint, which is like a unique digital representation of the track as one of the many reference songs contained in the apps database. , how these apps determine their audio fingerprints and how they match them to their potential POS counterparts, varies from app to app Currently, Shazam appears to be less flexible in determining analogies. in musical composition and vocal melodies compared to Google's latest uh music recognizer.

On the other hand, Google appears to be having trouble expanding its song database due to the complexity involved in its machine learning processes, although each of these different approaches reveals the strengths of each application. and weaknesses I am sure that at the rate at which artificial intelligence and computing capabilities are evolving at the moment, we will surely be able to count on increasingly accurate music recognition technologies in the future, an ultimate method for detecting musical plagiarism. I would like to mention it is also related to the music recognition technology we just mentioned and it would be uploading the music track to media streaming platforms like YouTube or Spotify which would normally check for copyright infringement during the upload processes. and publishing, and the advantage of using these platforms would be the large volume of published music contained in their databases that is constantly updated as users upload, users upload and publish their own work, so this AI song that I generated with audio is 100% free from being affected by any type of copyright. or intellectual property claim, for me the short answer is, I don't think so, although I didn't really get any red flags while investigating the song for plagiarism, there is definitely always a risk of creating something that is too similar to There's Something Else Than already exists and, as I mentioned above, the risk of plagiarizing existing music does not only apply to AI-generated music, but can also pertain to traditional music and artwork, for the moment, although I'm sure I can continue. work on this project without having to worry too much about any kind of copyright infringement if you are curious to know what the journey and outcome of this project can be, don't forget to stay tuned and subscribe to the channel and while you are there if you have appreciated the content so far.

Why not like the video and share it with someone who might also be interested? That's all for this video. Thank you so much for watching and I'll see you in the next one, take a walk along the creek beneath the willow dream and the banjo plays such a sweet tune, oh laughter, echoes of fields of gold, shared stories and old songs, all sing with the heart, so free, clap your hands, stomp your feet to the rhythm. of heartbeats echo through the trees in this melody of life we ​​find our way with the sun high in the sky until the end of the day we join the song Feel the joy it brings as we dance in the light that the morning s the life is on the canvas paint your scene in the hues of love and Evergreen feel the pulse of the earth beneath our feet through the silence The glow of twilight we still find Our Path NE the soft silver light of the moon until the dawn will gather around share the sound of the night sweet chorus as the stars begin to shine the music rings true Serene all the hearts under the sky so blue they spin lose your ground in the freedom felt in the magic of twilight Haze all our worries will melt away and We will find joy in every single day from the first light of dawn to where the fireflies play.