DiffRhythm: Generative Music (done quickly)

Channel:

NanoNomad

Subscribers:

2,810

Published on April 17, 2025 7:15:13 AM ● Video Link: https://www.youtube.com/watch?v=oYRSijJOBEA

Duration: 0:00

298 views

4

A look at DiffRhythm; a diffusion model for generative music. This one is FAST. I'm looking over the demos, sharing some installation notes, trying some demo generations, seeing what works and what doesn't, and trying to make a decent sounding tune. This is not a detailed tutorial.

DiffRhythm Huggingface demo - https://huggingface.co/spaces/ASLP-lab/DiffRhythm
DiffRhythm demo page - https://aslp-lab.github.io/DiffRhythm.github.io/
DiffRhythm GitHub - https://github.com/ASLP-lab/DiffRhythm

The modified scripts I mention in the video (not required, may not work, your mileage may vary, etc) - http://nanonomad.com/2025/04/17/diffrhythm-fast-full-length-song-generation/

[00:01] What am I doing? What is all of this
[00:46] DiffRhythm is fast
[01:02] There is an online demo, but...
[01:30] Let's listen and critique some of the demo songs
[08:10] Installation notes
[10:10] Generating the demo examples
[12:02] A few notes
[13:40] Trying some of my own random demos with questionable results
[14:20] Poor results attempting to generate sample-based music (rap)
[15:12] Better results with acoustic music
[15:30] Regenerating the folksy-country tune from the demo page
[16:40] Extending the lyrics and adjusting the timestamps of the song and results
[17:05] The adjusted infer.py and infer_utils.py

#generativeai #texttoaudio #diffrhythm