Name:Anil, Nagathil

School/Affiliation:Ruhr-Universität Bochum, Germany; McMaster University, Canada

Co-Authors:Benjamin Lentz, Giselle Mojica, Ian C. Bruce, Christiane Völter, and Rainer Martin

Virtual or In-person:In-person

Short Bio:

Anil Nagathil studied Electrical Engineering and Information Technology at Ruhr University Bochum (RUB), Germany, and the University of Birmingham, UK. He received the Dipl.-Ing. degree in 2009 and the Dr.-Ing. degree in 2016, both from RUB. From 2016 to 2023, he was a postdoctoral researcher at the Institute of Communication Acoustics (IKA) at RUB. Since 2023, he has been a Senior Research Scientist at IKA (currently on leave). From 2025 to 2027, he holds a Visiting Scholar position at McMaster University in Canada, funded by a Marie Skłodowska-Curie Global Fellowship. His research focuses on statistical and machine learning-based speech and audio signal processing, auditory and neural modeling, and EEG acquisition and analysis, for applications in hearing instruments.

Abstract:

Despite advances in cochlear implant (CI) technology, access to and enjoyment of music remains a significant challenge for most users of CIs. This has motivated research into music processing algorithms that emphasize vocals and rhythmic components relative to the accompaniment, thereby reducing the perceived complexity of music pieces. While such algorithms have shown to improve music enjoyment, they are often computationally demanding and thus they may not be suitable for real-time application in a CI. Therefore, in this work we propose a lightweight music processing algorithm based on the temporal adaptation mechanism of an auditory periphery model. The adaptation model enhances onsets while attenuating slowly-varying components, thus preserving essential vocal and rhythmic elements.

The proposed algorithm was evaluated in a listening experiment with 10 CI listeners for 16 selected pop/rock music pieces. Using a two-alternative forced choice test design, the listeners compared music signals processed with the proposed algorithm to corresponding unprocessed versions and to a baseline algorithm based on harmonic-percussive sound separation. The proposed algorithm was significantly preferred over the unprocessed condition with an average preference score of 67% (p = 0.031). The experiment revealed a non-significant difference between the proposed algorithm and the baseline method (51% average preference in favor of the proposed method, p = 0.844), despite a considerably lower computational complexity of the proposed algorithm.

This work is a promising step towards computationally efficient, real-time music processing with the aim of making music more accessible and enjoyable to CI listeners.