| Autor: |
Wygonski, John J., Sohie, Guy |
| Zdroj: |
Journal of the Acoustical Society of America; 1987, Vol. 81 Issue S1, pS78-S78, 1p |
| Abstrakt: |
Autoregressive moving average (ARMA) or pole-zero modeling techniques are known to provide better spectral estimation capabilities than autoregressive (AR) or all-pole methods such as linear prediction for nasalized speech and speech corrupted by noise. This paper describes the results of applying two ARMA modeling techniques to synthesized and natural segments of noisy and clean speech. The overdetermined normal equation (ODNE) and the extended-order singular value decomposition (SVD) methods of ARMA parameter estimation are compared to the covariance method of linear prediction and are evaluated using parametric distance measures in the cepstral domain. Use of the AR part of the ARMA model to estimate the envelope of noisy speech is demonstrated. Coefficient transformations among ARMA, AR, and cepstral parameters are also discussed. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
