Efficient Target Activity Detection based on Recurrent Neural Networks

TL;DR

This paper compares different neural network architectures for target activity detection in binaural listening devices, demonstrating that RNNs outperform FNNs in challenging acoustic environments.

Contribution

It introduces the use of RNNs, including LSTMs and GRUs, for TAD and evaluates their performance against FNNs in small network topologies.

Findings

RNNs outperform FNNs for TAD.

All RNN variants show improved detection accuracy.

Small RNNs are suitable for embedded systems.

Abstract

This paper addresses the problem of Target Activity Detection (TAD) for binaural listening devices. TAD denotes the problem of robustly detecting the activity of a target speaker in a harsh acoustic environment, which comprises interfering speakers and noise (cocktail party scenario). In previous work, it has been shown that employing a Feed-forward Neural Network (FNN) for detecting the target speaker activity is a promising approach to combine the advantage of different TAD features (used as network inputs). In this contribution, we exploit a larger context window for TAD and compare the performance of FNNs and Recurrent Neural Networks (RNNs) with an explicit focus on small network topologies as desirable for embedded acoustic signal processing systems. More specifically, the investigations include a comparison between three different types of RNNs, namely plain RNNs, Long Short-Term Memories, and Gated Recurrent Units. The results indicate that all versions of RNNs outperform FNNs for the task of TAD.