InvFlowFD: Reference-Free and Background-Set-Free Perceptual Music Quality Metric with Flow Matching Inversion

2026-08-04

AI Summary

Existing no-reference music quality assessment methods rely on clean audio background sets, which limits their generalizability. This work proposes InvFlowFD, the first framework capable of performing perceptual quality evaluation without any reference audio or background set. Leveraging a pretrained flow-matching model, InvFlowFD generates samples via unconditional flow inversion and compares their features against a prior distribution to effectively capture artificial distortions. Comprehensive quantitative experiments and large-scale human listening studies demonstrate that InvFlowFD aligns closely with human judgments regarding both audio distortion and generative model quality, significantly outperforming current metrics.

Abstract

Existing reference-free methods for evaluating music perceptual quality alleviate the need for paired noisy-clean data, but they still rely on a background set, which is used to compute aggregated statistics of clean audio samples. In this work, we propose a novel approach that eliminates this requirement, achieving background-set-free and reference-free quality estimation using only a pre-trained Flow Matching backbone. We demonstrate that unconditional Flow Matching inversion via simple Euler integration is sufficient to detect various artificial distortions and accurately rank music generation models against human perceptual judgments. We introduce InvFlowFD, which performs flow inversion and compares a group of inverted samples to the prior distribution. We evaluate our method against prior work, quantitatively and with a thorough human study. Results suggest that InvFlowFD is highly correlated with human perception of sound distortions, as well as generative models' quality, while being more flexible and less restrictive than existing metrics.

Problem

• reference-free
• background-set-free
• perceptual music quality
• audio distortion
• quality metric

Innovation

• reference-free
• background-set-free
• Flow Matching inversion
• perceptual quality metric
• music generation evaluation