The Hitachi-JHU DIHARD III System: Competitive End-to-End Neural Diarization and X-Vector Clustering Systems Combined by DOVER-Lap

TL;DR

This paper describes the Hitachi-JHU system for the DIHARD III challenge, combining five subsystems through DOVER-Lap to achieve competitive diarization error rates and secure second place in all tasks.

Contribution

The paper introduces an ensemble of five diverse diarization subsystems refined for optimal performance, combined using DOVER-Lap for improved accuracy.

Findings

Achieved diarization error rates of 11.58% and 14.09% in Track 1 full and core.

Achieved diarization error rates of 16.94% and 20.01% in Track 2 full and core.

Secured second place in all challenge tasks.

Abstract

This paper provides a detailed description of the Hitachi-JHU system that was submitted to the Third DIHARD Speech Diarization Challenge. The system outputs the ensemble results of the five subsystems: two x-vector-based subsystems, two end-to-end neural diarization-based subsystems, and one hybrid subsystem. We refine each system and all five subsystems become competitive and complementary. After the DOVER-Lap based system combination, it achieved diarization error rates of 11.58 % and 14.09 % in Track 1 full and core, and 16.94 % and 20.01 % in Track 2 full and core, respectively. With their results, we won second place in all the tasks of the challenge.