English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Transformer-based end-to-end neural speaker diarization (EEND) models utilizethe multi-head self-attention (SA) mechanism to enable accurate speaker labelprediction in overlapped speech regions. In this study, to enhance the trainingeffectiveness of SA-EEND models, we propose the use of auxiliary losses for theSA heads of the transformer layers. Specifically, we assume that the attentionweight matrices of an SA layer are redundant if their patterns are similar tothose of the identity matrix. We then explicitly constrain such matrices toexhibit specific speaker activity patterns relevant to voice activity detectionor overlapped speech detection tasks. Consequently, we expect the proposedauxiliary losses to guide the transformer layers to exhibit more diversepatterns in the attention weights, thereby reducing the assumed redundancies inthe SA heads. The effectiveness of the proposed method is demonstrated usingthe simulated and CALLHOME datasets for two-speaker diarization tasks, reducingthe diarization error rate of the conventional SA-EEND model by 32.58% and17.11%, respectively.

Improving Transformer-based End-to-End Speaker Diarization by Assigning Auxiliary Losses to Attention Heads