Optimal Bit Layering for Scalable Audio Compression Using Objective Audio Quality Metrics

Perceptual audio compression uses the idea of auditory masking to hide coding distortion. These auditory masking thresholds are obtained from mathematical models of the human ear. At low bitrates however, coding noise is significant and cannot be masked by the audio content. A perceptually scalable audio compression system, even at low bitrates, should generate a bitstream with layers of audio fidelity such that each layer improves the quality of the reconstructed audio that is just noticeable by the listener. In this paper we describe a low bitrate (8-64 kbps), scalable audio compression system which uses a residual weighted VQ algorithm to generate a scalable bitstream. To modify this bitstream so that it is perceptually scalable, we layer the different residual indices generated by the coder using objective audio quality metrics developed for evaluating highly impaired audio signals.