A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Microelectronics, School of Engineering and Technology

Thesis (M.Eng.) - Asian Institute of Technology, 2006

The question to increase the compression rate while the reproduction quality to be still acceptable leads to the term of perceptual audio compression. Audio perceptual coding based on a psychoacoustic model which using to eliminate the irrelevant information. Psychoacoustic model is a mathematical model of human perceptual of sound. The goal of model is to develop functions that are related to hearing sensations of human auditory system. For real-time applications, e.g. wireless speaker, headphones, or internet telephony, these applications do not require high audio quality but require low coding delay. To achieve this fashion, I designed a system with low complexity to reduce the computational times but the quality does not much different from audio CD. I used MDCT filter banks as subband coding. The advantage of MDCT is its perfected reconstruction. For quantizing. I compare the coding performance both full-range and narrow-range. A full-range quantization gave poor dequantized signals compare to narrow-range quantization. But full-range is acceptable for a low bit-rate requirement. Adding Huffman coding after quantization improves compression ratio less than 5%. It depends on the nature of data to compress. Huffman coding is useless in the case of less identical data. Thus I designed the system to check the identical data before perform Huffman coding. This codec gave the average compression ratio about 2.71: I or 62.5%. The number of bits per sample reduced to 5.99 bits per sample b'om 16 hits per sample