Speaker Verification Using Simple Temporal Features and Pitch Synchronous Cepstral Coefficients

TL;DR

This paper explores a speaker verification method using simple temporal features and pitch synchronous cepstral coefficients, achieving promising accuracy by combining these features.

Contribution

It introduces a novel feature set combining intra-pitch temporal information with pitch synchronous cepstral coefficients for speaker verification.

Findings

Achieved 91.04% accuracy on a 20-speaker database.

Combining features improved verification performance.

Analysis of misclassified speakers provided insights.

Abstract

Speaker verification is the process by which a speakers claim of identity is tested against a claimed speaker by his or her voice. Speaker verification is done by the use of some parameters (features) from the speakers voice which can be used to differentiate among many speakers. The efficiency of speaker verification system mainly depends on the feature set providing high inter-speaker variability and low intra-speaker variability. There are many methods used for speaker verification. Some systems use Mel Frequency Cepstral Coefficients as features (MFCCs), while others use Hidden Markov Models (HMM) based speaker recognition, Support Vector Machines (SVM), GMMs . In this paper simple intra-pitch temporal information in conjunction with pitch synchronous cepstral coefficients forms the feature set. The distinct feature of a speaker is determined from the steady state part of five cardinal spoken English vowels. The performance was found to be average when these features were used independently. But very encouraging results were observed when both features were combined to form a decision for speaker verification. For a database of twenty speakers of 100 utterances per speaker, an accuracy of 91.04% has been observed. The analysis of speakers whose recognition was incorrect is conducted and discussed .