MFA: TDNN with Multi-scale Frequency-channel Attention for Text-independent Speaker Verification with Short Utterances

TL;DR

This paper introduces MFA, a multi-scale frequency-channel attention mechanism integrated into TDNNs, significantly improving speaker verification accuracy, especially with short utterances, while reducing model complexity.

Contribution

The paper proposes a novel dual-path MFA mechanism for TDNNs, enhancing speaker feature extraction at multiple scales and improving performance under short utterance conditions.

Findings

Achieves state-of-the-art results on VoxCeleb dataset.

Reduces parameters and computational complexity.

Effective for short-utterance speaker verification.

Abstract

The time delay neural network (TDNN) represents one of the state-of-the-art of neural solutions to text-independent speaker verification. However, they require a large number of filters to capture the speaker characteristics at any local frequency region. In addition, the performance of such systems may degrade under short utterance scenarios. To address these issues, we propose a multi-scale frequency-channel attention (MFA), where we characterize speakers at different scales through a novel dual-path design which consists of a convolutional neural network and TDNN. We evaluate the proposed MFA on the VoxCeleb database and observe that the proposed framework with MFA can achieve state-of-the-art performance while reducing parameters and computation complexity. Further, the MFA mechanism is found to be effective for speaker verification with short test utterances.