Abstract

This study examines the robustness of voice biometrics when speech signals undergo audio codec transformations and sampling rate variations, conditions common in telecommunication networks. Speaker verification systems such as ECAPA-TDNN perform well on clean datasets, but their accuracy declines when low-bitrate codecs compress speech or when signals are resampled at reduced frequencies. In real-world deployments, systems adapt audio to bandwidth and storage limitations, often removing subtle acoustic details that support consistent speaker recognition. The research will analyse how codec settings and sampling rates, particularly those optimized for efficiency in bandwidth-limited systems, influence the stability of speaker embeddings. Instead of ranking codecs, the study will investigate how compression and resampling shape embedding quality and verification accuracy. To accomplish this, controlled experiments will be designed using a standardized speech dataset. Audio is systematically encoded, resampled, and decoded under multiple conditions, and the resulting signals are evaluated using cosine similarity based speaker embedding metrics, including Hit@k, mean comparisons-to-accept, and similarity regret. This study will contribute at both theoretical and practical levels. Theoretically, it will expand understanding of how bitrate and sampling distortions affect embedding behavior in modern verification systems. Practically, it will deliver recommendations for deploying voice biometrics in environments constrained by bandwidth or sampling rates. These recommendations will identify operating ranges where verification remains reliable and highlight caseswhere performance is likely to degrade. The findings will guide the development of more resilient pipelines for voice authentication in mobile and telephony applications.

Library of Congress Subject Headings

Biometric identification; Coding theory; Signal processing--Digital techniques; Sound--Recording and reproducing--Digital techniques; Electronic noise

Publication Date

12-2025

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Ioannis Karamitsos

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS