Fréchet Audio Distance: A Metric for Evaluating Music Enhancement Algorithms

By Kevin Kilgour, Mauricio Zuluaga, Dominik Roblek, Matthew Sharifi. arXiv.org, 2018

357 citations recorded by the source corpus.

We propose the Fr\'echet Audio Distance (FAD), a novel, reference-free evaluation metric for music enhancement algorithms. We demonstrate how typical evaluation metrics for speech enhancement and blind source separation can fail to accurately measure the perceived effect of a wide variety of distortions. As an alternative, we propose adapting the Fr\'echet Inception Distance (FID) metric used to evaluate generative image models to the audio domain. FAD is validated using a wide variety of artificial distortions and is compared to the signal based metrics signal to distortion ratio (SDR), cosine distance and magnitude L2 distance. We show that, with a correlation coefficient of 0.52, FAD correlates more closely with human perception than either SDR, cosine distance or magnitude L2 distance, with correlation coefficients of 0.39, -0.15 and -0.01 respectively.

Subfields: Computational Mechanics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/4766415f046760e16eb09dd7d04e3789d732dec1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

1. Yossi Adi
2. Wei Han
3. William Chan
4. Jiahui Yu
5. Jade Copet
6. Esa Rahtu
7. Alexandre D'efossez
8. Tal Remez

In Signal Processing

1. Gabriel Synnaeve
2. Yossi Adi
3. Yu Zhang
4. William Chan
5. Tal Remez
6. Wei Han
7. Jinyu Li
8. Quoc V. Le

Citations in the corpus

• Speech gesture generation from the trimodal context of text, audio, and speaker identity (2020, background): “…et al. 2016) and FID (Heusel et al. 2017) have recently become de facto evaluation metrics because they can measure the diversity of generated samples as well as their quality, and this concept was successfully applied to other generation problems (Kilgour et al. 2018; Unterthiner et al. 2019).”
• Simple and Controllable Music Generation (2023, methodology): “We report the FAD [Kilgour et al., 2018] using the official implementation in Tensorflow with the VGGish model 5 .”
• High Fidelity Speech Synthesis with Adversarial Networks (2019, neutral): “We note that Kilgour et al. (2019) proposed a similar metric, Fr´echet Audio Distance .”
• Taming Visually Guided Sound Generation (2021, methodology): “[30] adapted FID [27] to evaluate the fidelity of music enhancement algorithms.”
• Noise2Music: Text-conditioned Music Generation with Diffusion Models (2023, methodology): “We measure the quality of our text conditioned music generation model with two metrics: the Fr´echet Audio Distance (FAD) (Kilgour et al., 2018) which measures how the quality of generated audio clips compare to that of two benchmark datasets, e.g. the music split of AudioSet (Gem-meke et al.,…”