Google’s Text-To-Music Tool ‘MusicLM’ Available to the Public
At the start of January, Google unveiled MusicLM, an experimental AI tool capable of generating music based on text inputs. Like ChatGPT, Bard can generate long text output on a single command. Also, DALL-E has image generation abilities by the prompt.
Recently, Google announced that the text-to-music tool is now available for users.
Although the specific countries where the MusicLM tool is available to have yet to be mentioned by the company, interested individuals can join the waitlist to test it in the AI Test Kitchen.
The tool can be tested via web browsers, Android devices, and iPhones.
This tool is straightforward to use. Users can receive two different versions of a song by simply entering a prompt like “upbeat music for a party,” users can receive two different versions of a song. Users can listen to both versions and provide feedback by awarding a trophy to their preferred track, which helps enhance the model.
The company has collaborated with musicians like Dan Deacon to gather early feedback on the tool.
In published research on Github, Google shared a series of music samples 5 minutes duration generated using the MusicLM model with paragraph-long descriptions.
The research emphasizes that MusicLM uses a hierarchical sequence-to-sequence modeling approach for generating music based on specific conditions. It ensures consistent music output with a frequency of 24 kHz.
The research paper also mentions a “story mode” demo where the model can create music with specific time durations for each type of melody, like for meditation (0:00-0:15), waking up (0:15-0:30), running (0:30-0:45), and time to give 100% (0:45-0:60).