Speech Digitization Excitation Study.
STANFORD RESEARCH INST MENLO PARK CALIF
Pagination or Media Count:
This report presents the results of research on speech excitation signals in linear predictive coding LPC of speech. Initially, the research was pursued in two basic areas, the long-term memory approach pitch extraction and the short-term memory approach residual encoding. Later, all of the effort was devoted to the short-term memory approach, because this was considered the more promising. In the long-term memory area, the authors studied the problem of tracking a deconvolved glottal wave extracted from the residual signal of a linear prediction filter, and the feasibility of using formant-isolation techniques for time-domain pitch extraction. In the short-term memory area, the authors investigated various methods of encoding the residual signal to be used as an excitation signal of the LPC synthesizer. The major result of our research in the short-term memory area is the development of the residual-excited linear prediction RELP vocoder. The RELP vocoder offers the advantages of the linear predictive method with those of the voice-excited vocoder.
- Voice Communications