End-to-end Neural Diarization: From Transformer to Conformer

TL;DR

This paper introduces a Conformer-based end-to-end neural diarization system that improves speaker diarization accuracy by combining convolutional and Transformer architectures, with data augmentation and mixed training to address domain mismatch.

Contribution

The paper demonstrates that Conformer architecture enhances neural diarization performance and proposes a method to mitigate domain mismatch using mixed simulated and real data.

Findings

Conformer-based EEND outperforms Transformer-based models with 24% error reduction.

Data augmentation and convolutional subsampling improve EEND performance.

Mixing simulated and real data reduces domain mismatch, enhancing diarization accuracy.

Abstract

We propose a new end-to-end neural diarization (EEND) system that is based on Conformer, a recently proposed neural architecture that combines convolutional mappings and Transformer to model both local and global dependencies in speech. We first show that data augmentation and convolutional subsampling layers enhance the original self-attentive EEND in the Transformer-based EEND, and then Conformer gives an additional gain over the Transformer-based EEND. However, we notice that the Conformer-based EEND does not generalize as well from simulated to real conversation data as the Transformer-based model. This leads us to quantify the mismatch between simulated data and real speaker behavior in terms of temporal statistics reflecting turn-taking between speakers, and investigate its correlation with diarization error. By mixing simulated and real data in EEND training, we mitigate the mismatch further, with Conformer-based EEND achieving 24% error reduction over the baseline SA-EEND system, and 10% improvement over the best augmented Transformer-based system, on two-speaker CALLHOME data.