An Investigation of the Cross Correlation of Speech Signals,
Abstract
All information available in speech signals is used for classifying these signals according to the degree of connection between a given speech signal and a reference signal. An averaged cross correlation coefficient determined within the entire frequency-time range of existence of the test functions was adopted as the degree of connection. A sort of dynamic spectrograph was used for speech processing; from microphone M, the speech signal - via frequency corrector FC, band filters BF, and square-law detectors SD - was applied to 10 low-pass filters LPF (time constant, 10 m sec). The entire transmission channel had an averaging time of about 13 msec. Ten energy envelopes appeared at the LPF outputs. Time quantization was performed by switch Sw which scanned the channels at a rate of 800 Hz with a channel-interrogation time of 1.25 msec. Thus, a word said before the microphone M could be recorded in the storage of a digital computer; the latter could calculate the cross correlation factor between a said word and a previously-stored reference word. It was found that the above method can reliably recognize a limited number of words (five words pronounced by 7 men and 4 women were used in the experiments). (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 16, 1970
- Accession Number
- AD0717869
Entities
People
- B. N. Epifantsev
Organizations
- National Air and Space Intelligence Center