Quantized Perceptual Compressed Sensing for Audio Signal Compression

Compressed Sensing (CS) has been widely used for multimedia processing to reduce the number of the measurements required to acquire signals that are spare or compressible sparse in some basis. CS provides good quality of the restored signal even when the signal is not completely sparse and even also at high compression ratio. However, classical CS assumes that the measurements are real-valued and have infinite-bit precision that requires impractical hardware implementation. Quantized CS provides a solution to this problem. Different quantized compressed sensing techniques are developed in literature and recovery is possible even if only 1-bit is used for the quantization. In this work, we propose using different quantization vlaues, including 1-bit compressed sensing for perceptual audio signal compression in perceptual systems in order to clarify the effect of the quantization process on the achievable quality of audio signal. Mean Opinion Score (MOS) is used as metric to compare the perceived quality of audio signal of 1-bit CS and classical CS. Simulations results show that reasonable performance is achieved for different quantization CS compared to quantized classical CS.