Will Google’s AudioLM enrich the world of music?

Will Google’s AudioLM enrich the world of music?

San Francisco, Oct. 8, 2022

Google researchers have developed AudioLM, an AI system that generates similar-sounding songs and speech from just a few seconds of audio. AudioLM is not yet available to the public. The technology resembles speech models in that it predicts what should come next based on a prompt.

AudioLM can produce sounds such as speech and piano music that are almost indistinguishable from the original recordings. Google says it could speed up the AI training process for audio generation and eventually automatically generate music to accompany videos.

Unlike current systems that rely on text-based data, AudioLM requires no prior labeling or transcription. It can mimic a sound’s pitch, timbre, intensity and articulation, as well as background noise and speaker breath sounds.

Google has posted examples on its AudioLM website. A description of the framework is available on the Google AI Blog.