Google has made a significant stride in the field of music generation with the launch of MusicLM to generate music from text at Google I/O 2023 This innovative model generates high-quality music from text descriptions, enabling users to create customized and personalized soundscapes with ease.
MusicLM uses a hierarchical sequence-to-sequence modeling task to cast the process of conditional music generation. It generates music at 24 kHz, which remains consistent over several minutes, surpassing previous systems in audio quality and adherence to text description. The model can even be conditioned on both text and a melody and can transform whistled and hummed melodies to match the style described in a text caption.
To support future research in this field, Google has publicly released MusicCaps. This dataset comprises 5.5k music-text pairs with rich text descriptions provided by human experts.
Google has launched MusicLM in its AI Test Kitchen, where people can experience and provide feedback on some of Google’s latest AI technologies. The goal is to learn, improve, and innovate responsibly on AI together.
Everything in the test kitchen is work in progress, meant for early feedback, and will exist for a limited time. This space is for experimenting with different recipes and techniques, receiving feedback, and improving the technology.
According to Google, responsible progress in AI does not happen in isolation, and giving people the opportunity to experience the technology first hand is essential to learn and improve. Google plans to take things gradually, bringing in small sets of people at a time over the next few months to learn and improve the technology.
As Google launches MusicLM to generate music from text at Google I/O 2023, it is an exciting development in the field of music generation and is sure to have far-reaching implications for the future of music creation and composition.
Source: Google I/O Event