Noise2Music: Text-conditioned Music Generation with Diffusion Models

TL;DR

Noise2Music employs a two-stage diffusion model approach conditioned on text prompts to generate high-quality, semantically rich 30-second music clips, advancing text-to-music synthesis.

Contribution

It introduces a novel two-stage diffusion framework with intermediate representations for high-fidelity text-conditioned music generation.

Findings

Generated music reflects key text prompt elements like genre and mood

The approach grounds fine-grained semantics beyond basic features

High-quality 30-second music clips are produced from text prompts

Abstract

We introduce Noise2Music, where a series of diffusion models is trained to generate high-quality 30-second music clips from text prompts. Two types of diffusion models, a generator model, which generates an intermediate representation conditioned on text, and a cascader model, which generates high-fidelity audio conditioned on the intermediate representation and possibly the text, are trained and utilized in succession to generate high-fidelity music. We explore two options for the intermediate representation, one using a spectrogram and the other using audio with lower fidelity. We find that the generated audio is not only able to faithfully reflect key elements of the text prompt such as genre, tempo, instruments, mood, and era, but goes beyond to ground fine-grained semantics of the prompt. Pretrained large language models play a key role in this story -- they are used to generate paired text for the audio of the training set and to extract embeddings of the text prompts ingested by the diffusion models. Generated examples: https://google-research.github.io/noise2music