MIRNet: Learning multiple identities representations in overlapped speech

TL;DR

This paper introduces a deep learning method to extract multiple speaker identities from overlapped speech signals, enabling improved speaker verification and speech separation without needing reference features.

Contribution

It proposes a novel deep speaker representation network that extracts multiple speaker identities directly from overlapped speech using only identity labels, unlike traditional methods.

Findings

Effective in speaker verification tasks

Improves speech separation conditioned on speaker embeddings

Requires only speaker identity labels for training

Abstract

Many approaches can derive information about a single speaker's identity from the speech by learning to recognize consistent characteristics of acoustic parameters. However, it is challenging to determine identity information when there are multiple concurrent speakers in a given signal. In this paper, we propose a novel deep speaker representation strategy that can reliably extract multiple speaker identities from an overlapped speech. We design a network that can extract a high-level embedding that contains information about each speaker's identity from a given mixture. Unlike conventional approaches that need reference acoustic features for training, our proposed algorithm only requires the speaker identity labels of the overlapped speech segments. We demonstrate the effectiveness and usefulness of our algorithm in a speaker verification task and a speech separation system conditioned on the target speaker embeddings obtained through the proposed method.