Introduction

High quality audio signals have very high bandwidth; the industry standard is a 44.1 kHz sampling rate that covers the entire audible frequency range of human hearing (20 hz-20kHz). Each sample of the discrete signal is quantized using 16 bit linear PCM, resulting in a bit rate of 706.5 kbits/sec per channel. Audio compression algorithms must reduce this bit rate as much as possible with little or no loss in perceivable audio quality.

To accomplish this, we decided to implement a compression scheme that uses psychoacoustic modeling to determine which portions of the audio signal we remove without loss of sound quality to the human ear. The original signal is run through cosine modulated perfect reconstruction filter banks with 32 filters in each bank. The filter banks divide the signal into distinct frequency components and then the signal is quantized with a variable number of bits, which is based on the results of the psychoacoustic model. We have done analysis on this compressed version of the signal and by using different quantization schemes we can get 30 to 75 percent compression of the original signal. This difference is due to the overhead required to decode the quantized signal in each scheme.

We have gone through and then uncompressed these signals and compared their output to the original uncompressed signal to determine degradation in quality.

Here is a simplified block diagram of our scheme: