MultiSV: Dataset for Far-Field Multi-Channel Speaker Verification

TL;DR

This paper introduces MultiSV, a comprehensive multi-channel speaker verification dataset created through data simulation, enabling improved training and evaluation of multi-channel speaker verification systems in complex environments.

Contribution

The paper presents the MultiSV dataset, a new benchmark for multi-channel speaker verification, along with detailed data creation recipes and baseline system results.

Findings

MultiSV dataset facilitates training and evaluation in complex acoustic environments.

Baseline systems using neural network-based beamforming show promising results.

The dataset supports experiments with dereverberation, denoising, and speech enhancement.

Abstract

Motivated by unconsolidated data situation and the lack of a standard benchmark in the field, we complement our previous efforts and present a comprehensive corpus designed for training and evaluating text-independent multi-channel speaker verification systems. It can be readily used also for experiments with dereverberation, denoising, and speech enhancement. We tackled the ever-present problem of the lack of multi-channel training data by utilizing data simulation on top of clean parts of the Voxceleb dataset. The development and evaluation trials are based on a retransmitted Voices Obscured in Complex Environmental Settings (VOiCES) corpus, which we modified to provide multi-channel trials. We publish full recipes that create the dataset from public sources as the MultiSV corpus, and we provide results with two of our multi-channel speaker verification systems with neural network-based beamforming based either on predicting ideal binary masks or the more recent Conv-TasNet.