ConSinger: Efficient High-Fidelity Singing Voice Generation with Minimal Steps

TL;DR

ConSinger is a novel singing voice synthesis method that leverages the consistency model to produce high-fidelity singing voices efficiently with minimal inference steps, balancing quality and speed.

Contribution

This paper introduces ConSinger, a new SVS approach using the consistency model to achieve high-quality synthesis with fewer inference steps, improving efficiency over diffusion-based methods.

Findings

ConSinger produces high-fidelity singing voices comparable to baseline models.

It achieves faster inference with minimal steps while maintaining quality.

Experimental results demonstrate competitive speed and quality trade-offs.

Abstract

Singing voice synthesis (SVS) system is expected to generate high-fidelity singing voice from given music scores (lyrics, duration and pitch). Recently, diffusion models have performed well in this field. However, sacrificing inference speed to exchange with high-quality sample generation limits its application scenarios. In order to obtain high quality synthetic singing voice more efficiently, we propose a singing voice synthesis method based on the consistency model, ConSinger, to achieve high-fidelity singing voice synthesis with minimal steps. The model is trained by applying consistency constraint and the generation quality is greatly improved at the expense of a small amount of inference speed. Our experiments show that ConSinger is highly competitive with the baseline model in terms of generation speed and quality. Audio samples are available at https://keylxiao.github.io/consinger.