Skip to main navigation Skip to search Skip to main content

N for Parameter: Efficient Multi-Scale Neural Audio Super Resolution with GAN

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We introduce N-GAN, an end-to-end GAN architecture for neural audio super resolution that can accommodate multiple input sample rates. We refer to our approach as 'Wave-to-Wave' to distinguish it from the 'Wave-to-Spectrogram-to-Wave' and 'Wave-and-Spectrogram-to-Wave' approaches upon which the state-of-the-art results on this task are based. Our proposed 'Wave-to-Wave' architecture produces models that are orders of magnitude smaller than current state-of-the-art models whilst matching or exceeding their performance. In addition, our approach improves inference speed by at least 150% (2.5x speedup) over previous similarly performant models. We show that our model obtains state-of-the-art performance on a target sample rate of 48kHz and input sample rates of 8kHz, 16kHz and 24kHz.

Original languageEnglish
Title of host publicationECAI 2025 - 28th European Conference on Artificial Intelligence, including 14th Conference on Prestigious Applications of Intelligent Systems, PAIS 2025 - Proceedings
EditorsInes Lynce, Nello Murano, Mauro Vallati, Serena Villata, Federico Chesani, Michela Milano, Andrea Omicini, Mehdi Dastani
PublisherIOS Press BV
Pages313-321
Number of pages9
ISBN (Electronic)9781643686318
DOIs
Publication statusPublished - 21 Oct 2025
Event28th European Conference on Artificial Intelligence, ECAI 2025, including 14th Conference on Prestigious Applications of Intelligent Systems, PAIS 2025 - Bologna, Italy
Duration: 25 Oct 202530 Oct 2025

Publication series

NameFrontiers in Artificial Intelligence and Applications
Volume413
ISSN (Print)0922-6389
ISSN (Electronic)1879-8314

Conference

Conference28th European Conference on Artificial Intelligence, ECAI 2025, including 14th Conference on Prestigious Applications of Intelligent Systems, PAIS 2025
Country/TerritoryItaly
CityBologna
Period25/10/2530/10/25

Fingerprint

Dive into the research topics of 'N for Parameter: Efficient Multi-Scale Neural Audio Super Resolution with GAN'. Together they form a unique fingerprint.

Cite this