MusicLM is a GAMECHANGER for ML Text to Music Generation
HTML-код
- Опубликовано: 9 фев 2023
- Get notified of the free Python course on the home page at www.coursesfromnick.com
Sign up for the Full Stack course here and use RUclips50 to get 50% off:
www.coursesfromnick.com/bundl...
Hopefully you enjoyed this video.
💼 Find AWESOME ML Jobs: www.jobsfromnick.com
Links:
MusicLM Examples: google-research.github.io/sea...
Paper: arxiv.org/pdf/2301.11325.pdf
MusicCaps: www.kaggle.com/datasets/googl...
AudioLM Examples: google-research.github.io/sea...
Paper: arxiv.org/pdf/2209.03143.pdf
Oh, and don't forget to connect with me!
LinkedIn: bit.ly/324Epgo
Facebook: bit.ly/3mB1sZD
GitHub: bit.ly/3mDJllD
Patreon: bit.ly/2OCn3UW
Join the Discussion on Discord: bit.ly/3dQiZsV
Happy coding!
Nick 
Наука
![MusicLM Generates Music From Text [Paper Breakdown]](/img/1.gif)
Whatcha reckon guys? More ML 411s?
I think we need to do some more where we can have a run with the code.
Hi Nicholas, great video!
Do you know if there is a tool that exists that can generate new samples based on a text prompt?
Basically if I want a “TR-909 kick drum sample run through a boss ds-1 processed by an alesis nanoverb and finally sampled through an sp-1200.” And then the model generates a sample that I can use in my music.
Or something like “snare drum from despacito mixed with a tr-808 tom and metallica death magnetic snare” and it builds a sample based on those criteria?
If not, how hard would this be to develop a program that does something like this?
Yesssir
Some ppl are concerned that AI can replace human artists or musicians and trying to ban such models from being released / used / trained. I find this the dumbest move. If I love some musical band, I'll keep looking after them and purchase their new albums. I don't care if someone tries to rip them off to make his own music in the same style. The same goes for artists.
Yeah I’m with you on this, I think the idea of being able to generate stuff is great…just not at the expense of the original artists! Although, it’s the Wild West out there right now.
@@NicholasRenotte I think as we have factory made artifacts, still hand made ones are more valued and get higher price as well, I think same will go for authentic music or any other art form.
@@noob_techie2475 good point. The most screaming we hear now is coming from mediocre artists who couldn't withstand any form of competition, I guess. Real artists who are professionals, do not care
You're just awesome nick! Every morning I see one of your vd or shorts and I am charged up for the day and go to my job as a n excited data scientist 😌❤️
Hello, I just came across your channel this morning, and it has been 4 hours that I'm watching your videos and I just can't stop! You're really doing such good work.
Also, I have a really interesting project that I am working on, it is about generating a quiz from a text using NLP. But I'm really stuck right now. Can you help me please with some advice or an algorithm I could use and work with?
Also, it could be an interesting video idea! Maybe a 15mins challenge haha.
I'm planning to recreate MusicLM with the public datset, starting today
Oooooh, lmk how you go!!
Who knew tribal drums and flute got the party started 🪩
LOOOOOL, it’s a vibe
Just like stable diffusion has trouble generating images including sensical text, MusicLM has trouble generating words that make sense as it's not the main focus and there's not enough processing done around the texts of training songs.
Yeah I figured it had an acoustic resemblance to true lyrics but synthesising voices was a bridge too far. Can’t imagine that’s far away though!
it is millions of miles away from every day usefulness.
reminds me of texas instruments speech & spell for music
It's like you think and there is already something for it. Thanks @Nicholas Renotte for talking about it, this will help me a lot in my upcoming project 🤞🤞.
Hey!!! So glad you liked it, I’m working like crazy right now to much this stuff out. Pumped you checked it out 🙏
Love your videos, keep it up ❤
Real interesting! Another thing I heard recently is how copyright will be handled when bots are sampling real human artists 🤔
Hi Nicholas amazing video thanks for sharing, are you planning to do a video with the implementation in Python?
Hey hey I am new to deep/machine Learning.
Which one I need to learn first keras or Tansorflow .
Can you make a video of what transformer models are and give a simple example of how to build one. love your videos keep it up.
In the works 😉
fantastic breakdown ! excited to see the development , the examples sound great and love the story mode and conditional mode . however without code its unproven and unusable , like we are just trusting the results based on reputation that these are not the most cherry-picked examples / indicative of the results . if the dataset is ok to release then why would they not prove their model works . code is truth , until it is externally reproduced then its fluff .
Yeah agreed, it’s a tough one right? It’d be amazing if we could give it a crack!
hey nicholas, can you make a video on latest microsoft paper on VAll-E. IT is almost unbelievable it clones your voice with only 3 seconds voice sample. + Bonus if u can help implementing it too! as there exists very few implementations of the same with unclear instructions!
Will add it to the list, trying to catch up on LLMs right now but will look at it!
Death Metal Song ??? No way , I’m looking for more of a thrash metal/EDM but Death Metal will be a good start 😊
How do I use it ?
im confused, can we try it out or not?
How did you learn machine learning and deep learning
Is it able to have an api of musicml
i want them to be in a app
I don't think we should limit the creativity of humans in general just because someone is making money out of it. We should find a new paradigm for living that is based on other values than money now we have the chance.
programmers are the new rockstars!
🤘🏽
hahaha love your reactions
Do you have a phd in ML?🙄🙄
its like chatgpt, but 12 years ago. music AI is the area with results which sounds utterly untalented, keeping people unexited and unimpressed, for good reasons if you have had anything to do with music in your life, that means. lets say you need to create elevator muzac, there are other generative mass producing methods in digital music which do serve better than current transformer models, AI is SOOO behind in music - not just in sound quality, but in all in all convincing usability, maybe for a reason.
yeah you can make a AM radio quality of snoop dog which might you make giggle, you could use it as an effect such as autotune, but as all in all black box replacing musicians, show me an example. AI might be bad for all kind of creative bullshit jobs, with minimal creativity involved, or areas of craftmenship, as in applied arts, such as voice actors in gaming, or illustrators in the print sector. but music? go show us.
Good God the Micky Mouse editing and sfx of this video are annoying.
Job Offer! Can anyone help me install Google's LM music generator? I can pay for your time!
But...
It all sounds like 💩 not quite there yet, sounds like a machine.