Improved Speech Enhancement using Parallel MVDR Beamforming and Coherent-to-Diffuse Power Ratio Post-filtering

Sureshkumar Natarajan, Mohd Khair Hassan, Raja Kamil, Faisul Arif Ahmad, Syaril Azrad, June Francis Macleans, Hussna Elnoor Mohammed Abdalla, Nurbek P. Saparkhojayev, S. A. R. Al-Haddad

2026

Full text Open access: Hybrid Metadata (CSL-JSON)
DOI 10.47836/pjst.34.2.15
Data confidence 85 high
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Speech communication involves the exchange of information between two or more individuals; however, background noise and reverberation often degrade speech clarity and intelligibility. The impact of these degradations depends on factors such as the number, intensity, and spatial characteristics of noise sources, as well as reflections from surrounding surfaces. These effects pose significant challenges for applications including teleconferencing, hearing-aid devices, and human-machine voice interfaces. This study aims to address the combined effects of background noise and reverberation by proposing a speech enhancement framework that integrates Minimum Variance Distortionless Response (MVDR) beamforming with a Coherent-to-Diffuse Power Ratio (CDR) based post-filter. The methodology relies on parallel processing, where MVDR beamforming performs spatial noise suppression, while CDR values are estimated from microphone-domain signals and used to compute post-filter gains for suppressing residual diffuse noise and reverberation in the beamformer output. The proposed algorithm was evaluated over an input signal-to-noise ratio range from 0 dB to 40 dB and compared against the Integrated Sidelobe Cancellation Linear Prediction (ISCLP) baseline using four objective metrics: Perceptual Evaluation of Speech Quality (PESQ), Extended Short-Time Objective Intelligibility (ESTOI), Cepstral Distance (CD), and Weighted Spectral Slope Distance (WSS). The results demonstrate consistent improvements in PESQ and reductions in CD and WSS across all tested conditions, while gains in ESTOI remain modest. These findings indicate improved spectral fidelity and speech naturalness, highlighting the practical relevance of the proposed framework for real-time, low-complexity speech enhancement in noise and reverberation-prone communication systems.