Multi-Speaker Localization Using Convolutional Neural Network Trained with Noise

TL;DR

This paper introduces a CNN-based approach for multi-speaker localization trained with synthesized noise signals, demonstrating effectiveness in scenarios with two speakers and outperforming traditional methods.

Contribution

The paper presents a novel training method for CNNs in multi-speaker localization using synthesized noise, simplifying data preparation and improving performance.

Findings

Effective localization of two speakers demonstrated

Outperforms traditional steered response power method

Training with synthesized noise is viable

Abstract

The problem of multi-speaker localization is formulated as a multi-class multi-label classification problem, which is solved using a convolutional neural network (CNN) based source localization method. Utilizing the common assumption of disjoint speaker activities, we propose a novel method to train the CNN using synthesized noise signals. The proposed localization method is evaluated for two speakers and compared to a well-known steered response power method.