Google Experiments with AI Music Instrument Playground
Google has created a new artificial intelligence experiment, this time simulating 100 different musical instruments as if you were playing the keyboard.
The new musical instrument playground can simulate more than 100 instruments from around the world, including exotic instruments. Veena from India, Dizi from China and Mbira from Zimbabwe are represented.
Even a resonator guitar, but you have to order it by the brand name (Dobro) instead of the description.
Music LM can combine the sound of an instrument with adjectives like "melancholy" or "dramatic" to create a specific sound. For experienced musicians, the tool even has a sequencer that lets you layer and place up to four different instruments you've created—each with its own sound. According to Google, the starting point of this experiment was to explore a user interface based on Music LM, which inspires creativity and instrumental discovery around the world. You can try the "festive crystal flute" and imagine Lizzo playing a fun holiday song in December while you try out the playground yourself. Playground is a continuation of this MusicLM project, a generative AI model that can create high-quality music based on text instructions.
MusicLM was launched by Google in February 2023 and trained on a dataset of 280,000 hours of music to create songs from text prompts. According to the researchers, the model is built in such a way that it can turn a collection of sequentially written descriptions into a musical story based on existing melodies that can be whistled, hummed, sung or played on an instrument. Being able to play over 100 instruments means that Google trained MusicLM for hundreds of extra hours on each of these instruments.
Whether it's a crystal flute, mandolin or sitar, the algorithm has received hundreds of hours of playing these instruments to accurately recreate the sounds. Do the musicians who played these instruments know that their recordings were used in this way? Google hasn't released its MusicLM training model, so we're left to guess.
BW