Robust Speech Encoding System Under Noisy Environments

Speech encoding systems serve as fundamental components of wireless communication systems. Linear predictive coding (LPC)-based encoding is the most popular technique, primarily due to its simplicity. However, while LPC performs well in noise-free conditions, its performance deteriorates significantly in noisy environments. Over the years, numerous algorithms have been proposed to address real-time encoding issues, but their effectiveness diminishes considerably in the presence of background noise. In this work, we first analyze four state-of the-art enhancement techniques based on estimators of the signal magnitude spectrum. Subsequently, we propose a novel technique designed to enhance the robustness of speech encoding systems under noisy conditions. Our proposed system incorporates a background noise suppression module utilizing the soft mask estimator with a priori SNR uncertainty (SMPR). This module is integrated into the system prior to LPC-based encoding. The effectiveness of the proposed encoding system is evaluated using NOIZEUS speech data across four noise types and varying noise levels. Experimental results demonstrate significant improvements in encoding performance, highlighting the system's ability to maintain robustness in noisy environments.