Speakers Unembedded: Embedding-free Approach to Long-form Neural Diarization

TL;DR

This paper introduces a new embedding-free neural diarization framework that improves long-form audio speaker diarization accuracy and reduces computational complexity without relying on separate speaker embeddings.

Contribution

It proposes a novel approach applying EEND both locally and globally, eliminating the need for additional speaker embedding frameworks.

Findings

Achieves 13% and 10% relative DER reduction on Callhome and RT03-CTS datasets.

Provides marginal improvements over EEND-vector-clustering without extra speaker embeddings.

Discusses computational complexity and strategies for reducing processing times.

Abstract

End-to-end neural diarization (EEND) models offer significant improvements over traditional embedding-based Speaker Diarization (SD) approaches but falls short on generalizing to long-form audio with large number of speakers. EEND-vector-clustering method mitigates this by combining local EEND with global clustering of speaker embeddings from local windows, but this requires an additional speaker embedding framework alongside the EEND module. In this paper, we propose a novel framework applying EEND both locally and globally for long-form audio without separate speaker embeddings. This approach achieves significant relative DER reduction of 13% and 10% over the conventional 1-pass EEND on Callhome American English and RT03-CTS datasets respectively and marginal improvements over EEND-vector-clustering without the need for additional speaker embeddings. Furthermore, we discuss the computational complexity of our proposed framework and explore strategies for reducing processing times.