SWITCHED UNIFORM SCALAR QUANTIZATION ADAPTED TO MEAN AND VARIANCE FOR SPEECH CODING
Average power and variance are widely used in adaptation techniques in signal coding. A speech signal is usually assumed to be zero-mean; thus an average signal power is equal to the signal variance. However, this assumption is valid only for longer signals with a large number of samples. When the signal is divided into frames (especially if the number of samples within the frame is small) the speech signal within the frame may not be zero-mean. Hence, frame-by-frame adaptation to signal mean might be beneficial. A switched uniform scalar quantizer with adaptation to signal mean and variance is proposed in this paper. The analysis is performed for different frame lengths and the results are compared to an adaptive uniform quantizer that uses adaptation only to average signal power, showing an improved performance. Signal to quantization noise ratio (SQNR) is used as a performance measure.