Dual-Path Filter Network: Speaker-Aware Modeling for Speech Separation

TL;DR

This paper introduces the dual-path filter network (DPFN), a novel post-processing model for speech separation that leverages speaker identity information to enhance separation quality and avoids permutation-invariant training issues.

Contribution

The paper presents a new dual-path filter network that improves speech separation by incorporating speaker-aware modeling and is built upon DPRNN-TasNet, addressing permutation-invariance problems.

Findings

DPFN outperforms DPRNN-TasNet in speech separation tasks.

DPFN effectively utilizes speaker identity information for better separation.

The model avoids permutation-invariant training issues.

Abstract

Speech separation has been extensively studied to deal with the cocktail party problem in recent years. All related approaches can be divided into two categories: time-frequency domain methods and time domain methods. In addition, some methods try to generate speaker vectors to support source separation. In this study, we propose a new model called dual-path filter network (DPFN). Our model focuses on the post-processing of speech separation to improve speech separation performance. DPFN is composed of two parts: the speaker module and the separation module. First, the speaker module infers the identities of the speakers. Then, the separation module uses the speakers' information to extract the voices of individual speakers from the mixture. DPFN constructed based on DPRNN-TasNet is not only superior to DPRNN-TasNet, but also avoids the problem of permutation-invariant training (PIT).