Weekly Findings

New Incredible Text-To-Music Generation Model By Google

Image generated by Jacob Ferus

The following subjects are covered in this article:

• MusicLM — New Music Generation Model From Text
• YOLOPandas — Pandas Queries By AI In Jupyter Notebooks
• Deep Learning Interviews — Free Book About AI Interview Questions

MusicLM — New Music Generation Model From Text

About a month ago, in an earlier edition of weekly findings, I wrote about a new music generation model called Riffusion. This model was using a fine-tuned version of Stable Diffusion to transform text into spectrograms that were then converted to audio signals.

Now, Google has created a new model and it’s even better. It’s called MusicLM and has a variety of features such as:

• Audio generation from long-form text
• Ability to generate audio samples of several minutes
• Can use recorded humming in combination with text to generate music
• Given the same prompt, it can generate a variety of different sounds
• Generated signals are 24 kHz

The creators also publicly released a new high-quality text-to-music dataset that has been prepared by…