Towards perceptual metrics for audio quality assessment

Recent years have seen considerable advances in audio synthesis with deep generative models known as Neural Audio Synthesizers (NAS). However, the state-of-the-art is very difficult to quantify directly; different studies use different evaluation methodologies and metrics when reporting results, making a direct comparison to other systems difficult. Furthermore, in the case of objective metrics, the perceptual relevance and meaning of the reported metrics in most cases is unknown, prohibiting any conclusive insights with respect to the generated audio quality. This dissertation work focuses on: 1. Investigating how effective contemporary metrics are at capturing the audio quality of sounds generated from NAS systems, 2. Assessing the behavior of popularly used embedding distance measures and the embeddings that are used and, 3. Designing metrics that can measure the difference in audio quality between two signals, given that one of them is a reference signal. Thus, the proposed work will focus on developing objective metrics to provide comparable, reproducible and perceptually meaningful audio quality measurement strategies to help set the foundation for future NAS development.

Level thesis:degree_level

Doctoral

Department dc:contributor.department

Music

Grantor dc:publisher

Georgia Institute of Technology

Year dc:date.issued

2024

Author dc:creator

• Vinay, Ashvala

Advisor dc:contributor.advisor

• Lerch, Alexander

Committee members dc:contributor.committeemember

• Arthur, Claire
• Quinton, Elio
• von Coler, Henrik

• Audio Quality Assessment
• Sound Quality
• Signal Processing

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/1853/76868

OAI identifier oai:identifier

oai:repository.gatech.edu:1853/76868