DENSE: Dynamic Embedding Causal Target Speech Extraction

TL;DR

This paper introduces a dynamic embedding causal model for target speech extraction that improves real-time performance and captures contextual information, outperforming static embedding methods in challenging scenarios.

Contribution

The paper presents a novel autoregressive dynamic embedding approach for causal target speech extraction, addressing limitations of static embeddings.

Findings

Enhanced STOI and SDR metrics in experiments

Effective real-time, frame-level extraction demonstrated

Outperforms static embedding models in challenging scenarios

Abstract

Target speech extraction (TSE) focuses on extracting the speech of a specific target speaker from a mixture of signals. Existing TSE models typically utilize static embeddings as conditions for extracting the target speaker's voice. However, the static embeddings often fail to capture the contextual information of the extracted speech signal, which may limit the model's performance. We propose a novel dynamic embedding causal target speech extraction model to address this limitation. Our approach incorporates an autoregressive mechanism to generate context-dependent embeddings based on the extracted speech, enabling real-time, frame-level extraction. Experimental results demonstrate that the proposed model enhances short-time objective intelligibility (STOI) and signal-to-distortion ratio (SDR), offering a promising solution for target speech extraction in challenging scenarios.