Spectral Methods for Single Channel Speech Enhancement in Multi-Source Environment

Abstract

Speech communication for both humans and automatic devices can be negatively impacted by background noise, which is common in real environments. Among many techniques, speech separation using a single microphone is the most desirable from an application standpoint. The resulting monaural speech separation problem has been a central one in speech processing for several decades. However, its success has been limited thus far. This research presents work that develops speech separation systems using combinations of T-F masking, DNNs, and model-based reconstruction. The aim of each system is to improve the perceptual quality of the speech estimates. The performance of many speech processing applications is severely degraded when both noise and reverberation are present. The proposed solution has been tested in the simulation environment and based on the simulation result, it is observed that the speech enhancement can easily be performed through the integration of the solution. This research suggests two staged noise reducing systems in order to reduce the background noise through a single microphone recording in a low-SNR based on ideal binary masking and Wiener filter. It has two stages. Firstly, for background noise reduction, a Wiener filter with an enhanced SNR is utilised on noisy speech. Secondly, IBM is calculated in each time–frequency channel through utilisation of the pre–processed speech from the first stage and the matching of the time–frequency channels to a pre-selected threshold in order to minimise residual noise. These channels meeting the threshold requirement are conserved while all the other ones are attenuated.