No ratings yet.

After researching the consequences of AI voice models on, for instance, Drake and The Weeknd, I was curious about these models. I’m sure most of you have seen a video about it on Tiktok or Instagram about maybe a cartoon character or even toothbrushes that use popular songs and make the AI covers:

Then I wondered if I could make my “not-so-beautiful voice” sing like Bruno Mars. So, I had to train my own AI voice model. After some research online it comes down to having at least 10 minutes of clear vocal audio that you insert into the AI model to train it to your voice successfully. A clear vocal take is pretty important for it to work. After this, you are led to an online Google Drive that lays out all the steps to successfully train your voice model. I tried to make my voice sound like Bruno Mars’ song “When I Was Your Man”. You extract the instrumental and the vocal with an online AI tool and make the trained model replace Bruno’s voice. Then the result…

I will spare you the result, but it is scary good. For most of the song, it actually sounds like I sang it. You can still definitely hear it’s a generated voice but if I let this play for my grandparents, they would probably think they have a new star in their presence. It’s very surreal that you can just make your voice sound like a very popular artist.

Other than the fun side of creating this, this could also very well be used for the wrong reasons. I could copy stuff from less-known artists and use it for my benefit because I said in my last blog post the current copyright technology doesn’t recognize AI-generated voices. This could also lead to unethical use of AI.

Try it for yourself, it’s a fun experiment that if used properly could definitely bring some laughs.

No ratings yet.

The recent developments of Generative AI within the music industry have been really interesting to me. As a BIM student and very active music hobbyist, I’ve seen technology influence a lot of aspects of the music-making process. It started with AI programs that could find sounds within your music library that would fit in with the track you were making. Finding sounds was never this easy. This was quite an innocent addition if you compare it to the possibilities that are available now.

One of them is generative AI Voice models, with these models it was possible to recreate any song with any voice. At first, this sounded very cool, but this quickly turned into several controversies. People were copying voices from major artists and layering the vocals on top of instrumentals they made or generated. The songs were so believable that unreleased material could not be separated from generated material. See for instance the following song that used the voice of popular artist Drake:

Copyright is a big issue with these songs. A copyright system works by recognizing existing patterns, but with AI, these are all new patterns, so copyright claims do not work against these generated AI songs. See for instance the following copyright infringement article about popular artists The Weeknd and Drake:

https://www.theverge.com/2023/5/1/23703087/ai-drake-the-weeknd-music-copyright-legal-battle-right-of-publicity

These cases will create some interesting and challenging questions. What if generated Drake songs divert revenue from the actual Drake? What if his voice is being used for malinformation? Other than the problems these AI voice models bring It probably won’t be very long till it’s possible to make believable full instrumentals with vocals singing lyrics that are all fully generated by AI. Music what first was this creative outlet for people becomes an industry where you have to be skeptical about the validity of songs. The fun and cool parts of the music you listen to might just be fully generated now.