Perceptual quantization using JNLD thresholds

In perceptual audio encoders the quantization noise is shaped according to the masking threshold of the signal obtained from the psycho-acoustic model. This approach of quantization minimizes the quantization-noise to masking-thresholds ratio (NMR). However, degradation in the perceived quality of the quantized signal may also result from the modulation difference of the temporal envelopes of the original and quantized signals. A joint cost function comprising masking thresholds and just noticeable level difference (JNLD) thresholds is proposed for the quantizer. This cost function provides better control over NMR, probability of detection of quantization error and modulation difference. Experiments with MPEG-2 AAC encoder have shown that 82% of a wide variety of test-signals chosen from EBU-SQAM improved in perceptual quality with the modified algorithm.