Abstract
dc:description.abstractReservoir Computing (RC) presents a promising pathway for efficient temporal signal processing, yet its application in audio has largely been confined to the role of a classifier. This perpetuates reliance on the conventional, multi-stage audio processing pipeline, where computationally expensive time-frequency transformations like Mel-Frequency Cepstral Coefficients (MFCCs) remain a bottleneck. This inefficiency limits the deployment of audio systems in real-time, low-power scenarios. In this work, we fundamentally redefine the use of RC for audio by introducing a unified framework that collapses the entire traditional pipeline. We demonstrate that a single reservoir can be configured to function not only as a classifier but also as a powerful feature extractor, operating directly on raw audio waveforms in the time domain. Specifically, we show the reservoir’s high-dimensional dynamics can be trained to mimic conventional MFCC extraction and generate equivalent features without the need for domain transformations. Our results confirm that this integrated approach successfully enables end-to-end time-domain processing from raw audio to classification, significantly reducing computational complexity and latency. This work establishes RC as a versatile and efficient alternative for audio processing, paving the way for its integration into next-generation technologies for embedded systems, and real-time speech interfaces.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of York
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sebastian, Rinku
- Advisors dc:contributor.advisor
-
- Trefzer, Martin
- O'Keefe, Simon
- ORCID person:identifier.orcid
Rights
- Language dc:language.dcterms_iso639_2
- eng