Statistical estimators of a periodically correlated random process for a voiced speech signal
Lesya B. Chorna · The Journal of the Acoustical Society of America · 2003
A stochastic model in the form of a periodically correlated random process (PCRP) was applied to a voiced speech signal. In such a signal, the amplitudes and phases of the harmonic components varied randomly; therefore, a stochastic model of the signal was appropriate. A description of the signal in terms of a stationary random process allowed an analysis of the spectral density of the oscillations, but the phase (temporal) structure remained unconsidered. To reveal this structure, a PCRP was used. The speech signal was processed by the co-phase analysis in the time domain and the component analysis in the spectral domain. The algorithms were based on a specific property of the PCRP that the samples selected with a period of correlation formed a series of related stationary sequences. Each sequence was then analyzed using the theory of stationary processes. This technique was used in a study of 70 recordings of the vowel /a/. The recordings were obtained from people with a normal heart rate and provoked arrhythmia. Statistical estimators of the vocal signal calculated on the basis of the PCRP model were statistically different for the normal and arrhythmic heart rates.