ABSTRACT

Code excited linear prediction (CELP) coding methods perform well at bit rates near 5 kbit/s and above. However, as system requirements lower the bit rate below that level, the quality of CELP output speech declines precipitously. This is primarily due to the fact that not enough bits are available to represent a sufficient number of codes for the excitation. Or viewed differently, the coder expends too many bits attempting to copy perceptually unimportant time-domain details of the excitation signal.