With the rising interest of digital recording and sound storage it is important to be able to reduce the number of bits of data as much as possible while still preserving the highest degree of fidelity. Two coding schemes were tested for their usefulness in this context: Huffman coding, which is an information-preserving coding scheme, and the system of floating-point coding which is roughly related to ADPCM coding and to predictive coding. This latter scheme is capable of reducing 22-bit samples to 13 bits with very little perceptible loss of quality. Discrimination tests were performed to determine the fidelity of the coded signal with the somewhat striking result that much more error can be tolerated in the midfrequencies (1 kHz to 6 kHz) than at either the low or high ends of the audio spectrum. A tentative explanation is offered for this effect based on masking phenomena and loudness summation.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.