A Comparative Re-Assessment of Feature Extractors for Deep Speaker Embeddings

TL;DR

This paper evaluates 14 different feature extraction methods for deep speaker embeddings, finding that certain spectral and noise suppression techniques outperform traditional MFCCs, reducing error rates significantly.

Contribution

It provides the first extensive comparison of alternative feature extractors for deep speaker verification, highlighting promising techniques beyond MFCCs.

Findings

Spectral centroid features improve EER by up to 16.3% on VoxCeleb.

Group delay features reduce EER by significant margins.

Integrated noise suppression enhances speaker verification accuracy.

Abstract

Modern automatic speaker verification relies largely on deep neural networks (DNNs) trained on mel-frequency cepstral coefficient (MFCC) features. While there are alternative feature extraction methods based on phase, prosody and long-term temporal operations, they have not been extensively studied with DNN-based methods. We aim to fill this gap by providing extensive re-assessment of 14 feature extractors on VoxCeleb and SITW datasets. Our findings reveal that features equipped with techniques such as spectral centroids, group delay function, and integrated noise suppression provide promising alternatives to MFCCs for deep speaker embeddings extraction. Experimental results demonstrate up to 16.3\% (VoxCeleb) and 25.1\% (SITW) relative decrease in equal error rate (EER) to the baseline.