Speech Compression by Spectral Decomposition
A. Arcese, A. J. Goldberg · The Journal of the Acoustical Society of America · 1974
In this paper, we consider compressing speech using diagonalization transformations which include predictive coding as a special case. The algorithm determines the autocorrelation matrix pitch synchronously, producing a circular Toeplitz matrix under stationarity assumptions. Next, we find a diagonalization transformation involving both the eigenvalues, which for this matrix is the spectrum, and the eigenvectors, which depend only on the dimension of the matrix and are independent of the eigenvalues. This transformation then operates on the speech waveform, producing a random sequence that is then quantized and transmitted along with parameters characterizing the autocorrelation matrix. The synthesizer calculates the inverse transform exactly by Fourier transforms and then reconstructs the waveform. This paper presents results illustrating these processing operations.