RECOGNITION OF SPOKEN WORD ENSEMBLES WITH THE AID OF A DIGITAL COMPUTER,
FOREIGN TECHNOLOGY DIV WRIGHT-PATTERSON AFB OHIO
Pagination or Media Count:
It is shown that the attributes of speech, which are naturally related to differential attributes of phonemes, articulation and, particularly, acoustic features, must be affected to a much smaller extent than the speech signal itself by the features of individual human pronunciation and partly also by context. Then the speech signal is construed as consisting of a number of segments, each characterized by one or several attributes such as the presence of fundamental tone, the noise component high density of zeros, etc. The introduction of segments facilitates the computer recognition of speech since then a spoken word can be broken down into a sequence of segments, recognizable on a probabilistic basis. In this connection, certain attributes of speech signals were experimentally investigated by having 20 speakers of both sexes pronounce 10 letters of the Cyrillic alphabet each. The attributes of the sound signal were isolated on the basis of time-frequency pattern matching and represented in binary form.
- Voice Communications