Computer Identification of Phonemes in Continuous Speech.
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OHIO SCHOOL OF ENGINEERING
Pagination or Media Count:
An approach to computer identification of phoneme-sized sound elements in continuous speech is presented. Using a tape recorder as an input device, analog speech signals are digitized and fast Fourier transformed. The amplitudes of the transformed signals are combined in a logarithmic manner and the resulting 16 channels of data used as the input to a phoneme recognition scheme. A running crosscorrelation is performed with 47 stored phoneme prototypes. The locations of possible phonemes are determined and one set of possible phonemes selected. The procedure was performed on three replicates of 45 isolated words and on three replicates of ten sentences.
- Voice Communications