SqueezeComposer: Temporal Speed-up is A Simple Trick for Long-form Music Generation

Abstract

Composing coherent long-form music remains a significant challenge due to the complexity of modeling long-range dependencies and the prohibitive memory and computational requirements associated with lengthy audio representations. In this work, we propose a simple yet powerful trick: we assume that AI models can understand and generate time-accelerated (speeded-up) audio at rates such as 2x, 4x, or even 8x. By first generating a high-speed version of the music, we greatly reduce the temporal length and resource requirements, making it feasible to handle long-form music that would otherwise exceed memory or computational limits. The generated audio is then restored to its original speed, recovering the full temporal structure. This temporal speed-up and slow-down strategy naturally follows the principle of hierarchical generation from abstract to detailed content, and can be conveniently applied to existing music generation models to enable long-form music generation. We instantiate this idea in SqueezeComposer, a framework that employs diffusion models for generation in the accelerated domain and refinement in the restored domain. We validate the effectiveness of this approach on two tasks: long-form music generation, which evaluates temporal-wise control (including continuation, completion, and generation from scratch), and whole-song singing accompaniment generation, which evaluates track-wise control. Experimental results demonstrate that our simple temporal speed-up trick enables efficient, scalable, and high-quality long-form music generation.

**Figure 1:** (A) Audio is compressed via time-scale modification, generated in a compact domain for various tasks (e.g., long-form composition, accompaniment), and then expanded and refined to the original resolution. (B) Training pipeline for temporal speed-up and restoration. The input audio is speeded up and processed through a two-stage pipeline: CNN generates a prior condition, then DiT refines it to produce high-quality restored audio. The accelerated audio maintains the original sampling rate, ensuring vocoder compatibility. Training uses MSE loss for CNN prior generation and diffusion loss for DiT refinement.

**Figure 2:** Overview of the SqueezeComposer framework. (A) Long-form Music Composing: temporal speed-up enables efficient abstract-level generation using DiT with masking strategies (scratch, completion, continuation). (B) Whole-song Singing Accompaniment Generation: semantic-to-prior mapping followed by DiT refinement for harmonious accompaniment.

1. Squeeze and Restore

We propose a simple yet effective method that leverages AI's ability to understand and reconstruct accelerated music (such as 2x, 4x, or 8x) to reduce duration and resource requirements, enabling long-form music generation while restoring the original speed at the end. Below are reconstruction samples demonstrating the effectiveness of our approach at different acceleration rates.

	4x Speeding-up	Ground-truth	Reconstruction
Sample 1

Sample 2

Sample 3

	8x Speeding-up	Ground-truth	Reconstruction
Sample 1

Sample 2

Sample 3

2. Whole-song Accompaniment Generation

RandSong is randomly selecting an accompaniment from the accompaniment library to pair with the input vocal, FastSAG generates by cutting into 10-second segments, and SqueezeComposer is our proposed method that can generate whole-song accompaniment more quickly, coherently, and harmoniously.

2.1 The following samples are singing accompaniment generation. We compare our results with some baseline models.