Attentive Merging of Hidden Embeddings from Pre-trained Speech Model for Anti-spoofing Detection

TL;DR

This paper explores how hierarchical embeddings from a pre-trained speech model can be effectively merged using an attentive method to improve anti-spoofing detection accuracy, achieving state-of-the-art results.

Contribution

It introduces an attentive merging technique for multi-layer embeddings of WavLM, enhancing anti-spoofing detection performance and revealing the importance of early transformer layers.

Findings

Achieved EERs of 0.65%, 3.50%, and 3.19% on three ASVspoof datasets.

Early hidden layers of WavLM are highly effective for anti-spoofing.

Partial pre-trained models can maintain high performance with improved efficiency.

Abstract

Self-supervised learning (SSL) speech representation models, trained on large speech corpora, have demonstrated effectiveness in extracting hierarchical speech embeddings through multiple transformer layers. However, the behavior of these embeddings in specific tasks remains uncertain. This paper investigates the multi-layer behavior of the WavLM model in anti-spoofing and proposes an attentive merging method to leverage the hierarchical hidden embeddings. Results demonstrate the feasibility of fine-tuning WavLM to achieve the best equal error rate (EER) of 0.65%, 3.50%, and 3.19% on the ASVspoof 2019LA, 2021LA, and 2021DF evaluation sets, respectively. Notably, We find that the early hidden transformer layers of the WavLM large model contribute significantly to anti-spoofing task, enabling computational efficiency by utilizing a partial pre-trained model.