Google has unveiled a new AI tool that can generate music in any genre from
text prompts or even transform a hummed or whistled melody into different
instruments.
MusicLM, the technology behind this feat, is a text-to-music generation
system that analyzes text to decipher the scale and complexity of the
composition.
Keunwoo Choi
@keunwoochoi
whoa, this is bigger than ChatGPT to me.
google almost solved music generation, i’d say.
https://google-research.github.io/seanet/musiclm/examples/
The model was trained on 280,000 hours of music, enabling it to generate
coherent songs and capture nuances such as mood, melody, and instruments. In
addition, MusicLM can create a sort of melodic narrative by taking several
text descriptions and generate audio “played” by a specific instrument in a
certain style.
However, the Google researchers acknowledge the ethical challenges posed by
the system, such as the risk of incorporating copyrighted material from
training data into generated songs. During experiments, about 1% of the music
generated was found to be directly replicated from the training songs. The
authors emphasize the need for more work to tackle these risks associated with
music generation.
MusicLM offers exciting possibilities for music creation, but its use also
brings up ethical considerations. Google’s research team is working to address
these challenges and ensure that MusicLM can be used in a responsible and
sustainable manner.
