A Novel Speech Compression Technique Using Optimized Wavelet Transform to Improve the Quality of Auditory Perception Under Low SNR Conditions

Abstract

Speech compression in poor environment, where the signal energy is weak due to acoustical disturbances, can improve the efficiency of transmission while reducing the bandwidth if intelligibility and/or quality can be preserved by selecting appropriate energy based wavelets. We propose an Optimized Wavelet Transform (OWT) to improve speech perception by incorporating masking techniques in the algorithm to reduce the noise effect. Adaptive Wavelet Selection followed by optimized quantization exploit a robust Dynamic Dictionary Scheme (DDS) to perform efficient compression while preserving speech intelligibility and perceptual quality. An additional lossless coding technique inevitably increases the compression ratio while preserving the quality of the signal. Finally, decompressing the compressed signal undergoes tonal and noise masking by applying a global threshold based on Sub-Band Perceptual Factor (SBPF) and Perceptual Entropy (PE), which improves the quality of the signal. Performance of the proposed algorithm is obtained in terms of Normalized Root-Mean Square Error (NRMSE), Compression Ratio (CR), Performance Evaluation of Speech Quality (PESQ), Re-construction Distortion Length (RDL), Signal to Noise Ratio (SNR) for various voiced and unvoiced signals recorded in low SNR conditions. All the signals are derived from NOIZEUS data base and some samples are recorded and normalized to operate at sampling frequency of 8KHz.