Google created the MusicLM neural network, which generates music of any genre based on a textual description
Google has introduced a new generative neural network called MusicLM. This is a new AI model that can create high-quality music based on a text description. 280 hours of musical compositions were used to train the neural network. As a result, it is able to generate music of various genres based on an abstract description such as “soothing violin melody” and so on. Also available is the generation of music based on a picture and a description for it.
MusicLM creates music at a sample rate of 24 kHz in any genre and with any instruments specified in the description. The project page contains examples of compositions generated according to a detailed description. It’s an “upbeat arcade soundtrack with guitar riffs”, “slow reggae with bass, drums and electric guitar”. Some examples are surprisingly high-quality and even have vocals. Others have a lot of noise and distorted sounds. However, the results are still impressive. MusicLM is still under development, it will not be possible to personally test the capabilities of AI.
It should be noted that there is already a music generation project based on the Stable Diffusion neural network using visual spectrograms. And this AI model is available for public and commercial use.
Against the background of the success of the popular neural network Midjourney for generating images, it cannot be ruled out that in the future neural networks will partially replace people in the creation of audio-visual content.
