Data Compression Of Speech Signals
John Burt Foster, Tzung-Kwang Wang · 1991
The research presented in this paper is a continuation of previous research [J. Foster, T. Wangl which proposed an uniform amplitude coder as an alternative to the traditional time domain coders such as PCM, DM. The encoder of Time Code Modulation (TCM) transmits the quantized intervals corresponding to the time at which the input signal passed through uniform spaced amplitude levels (of a predetermined step size). The decoder reconstructs the signal by using linear interpolation in the amplitude domain (instead of time domain). Allowing for a variable rate transmission rate, TCM systems can effectively be used with packet switched networks. From the results of past research the TCM coder has shown promising performance. When sampling the analog voice signal (12 bit amplitude resolution) at a frequency of fs = 32KHz the resulting Signal-to-Noise (quantization noise) Ratio is 13.18 dB (using a step size = 32). The corresponding transmission rate is 20% to 30% of the sampling rate. As will be shown the sampling frequency and step size are the two dominate factors that affect TCM performance; the higher the sampling frequency and the smaller the step size a higher SNR can be achieved by TCM, but at the same time, the transmission rate will increases. In order to determine the optima value of each factor, a comparison was conducted on the sampling theorems between time and amplitude domains. Several parallels can be found between Nyquist based PCM and TCM systems. With uniform time window sampling the bandlimited nature was imposed upon the frequency of the input signal. How is the Nyquist based theorem for the uniform amplitude window sampling defined? Concerning the interpolation process at the decoder, is there an low-pass filter in the amplitude domain with parallels low-pass filters in the frequency domain? Since time and frequency have an inverse relationship, what is the relationship between amplitude and inverse-amplitude? The greatest gain of such a system may be found in its capability to change the l/amplitude (frequency) characteristic of the signals. By scaling the time of an original