|
KMID : 0368019910140030937
|
|
Journal of Soonchunhyang University
1991 Volume.14 No. 3 p.937 ~ p.942
|
|
|
Speech Recognition of Korean Word using LPC Cepstrum Coefficients
|
|
|
|
|
Abstract
|
|
|
This paper discusses speech recognition of Korean word by the LPC technique. Such a procedure is important in problems of man－machine communication.
The voice signals whose sampling frequency is 10 ㎑ is digitized and applied by 25.6 ㎳ of Hamming window to calculate autocorrelation {Ri}. Then the linear predictor coefficient {ai} is obtained by the Durbin´s algorithm at predictor order p=10 and the cepstrum coefficient {ci} is transformed in the base of {ai}.
Reference templates are created to allow maximum 10 words by clustering algorithm. DTW algorithm is used to compare similarity between reference templates and input words.
in analysis, this technique has required a longtime and the complicated hardware system but it has computed accurate results than any other method.
In recohnition, Euclidean and Chebyshev error rate are 11.6%m, 12.1% respectively.
If we improve the hardware system and apply pole－zero model, we can get more accurate results.
|
|
|
KEYWORD
|
|
|
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|