Speech Enhancement Results Evaluation Framework

The Speech Enhancement Results Evaluation Framework serves as a crucial tool for assessing speech quality and facilitating corresponding improvements. This framework incorporates four distinct speech quality evaluation methodologies: Signal-to-Noise Ratio (SNR) measuring overall noise reduction, Segmental SNR (segSNR) providing frame-by-frame analysis, Log Spectral Distance (LSD) assessing spectral accuracy, and Perceptual Evaluation of Speech Quality (PESQ) evaluating human-perceived quality. These complementary metrics enable comprehensive analysis of speech enhancement performance and identify areas for algorithmic optimization. The framework includes MATLAB implementation files (.m files) containing functions for generating speech files with user-defined SNR levels through controlled additive noise injection. This functionality allows systematic testing under various noise conditions by implementing SNR calibration algorithms that adjust noise power relative to clean speech signals. Designed with modular architecture, this evaluation framework not only serves speech enhancement applications but can also be adapted for speech quality assessment in other domains such as speech coding, audio restoration, and telecommunications systems. The implementation provides standardized interfaces for integrating custom speech processing algorithms and extends support for additional evaluation metrics through modular function design.