Vocoders compress speech by estimating model parameters at a given transmission rate over an analysis window, assuming that speech is stationary within this window. In this paper, the limits of this assumption are explored with regard to the spectral envelope parameters in the form of line spectral frequency (LSF) parameters. It is shown that all LSF parameters have considerable variations over time, regardless of LSF vector extraction and transmission rates. LSF track variations are investigated through oversampling and are shown to contain high frequency variations above the frequency corresponding to the LSF vector transmission rate. An anti-aliasing filter with cut-off frequency adequate for the chosen LSF vector transmission rate is proposed to alleviate possible spectral overlapping of the LSF parameter spectra. It is confirmed, through experiments, that the proposed method offers an advantage over the classic LSF extraction method with respect to quantisation shown by bit savings of typically 10 to 15%.
[1]
Thomas Eriksson,et al.
Low-rate quantization of spectrum parameters
,
2000,
2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100).
[2]
W. Bastiaan Kleijn,et al.
Spectral dynamics is more important than spectral distortion
,
1995,
ICASSP.
[3]
Ahmet M. Kondoz,et al.
Efficient parameter quantisation for 2.4/1.2 kb/s split-band LPC coding
,
2000,
2000 IEEE Workshop on Speech Coding. Proceedings. Meeting the Challenges of the New Millennium (Cat. No.00EX421).
[4]
F. Itakura.
Line spectrum representation of linear predictor coefficients of speech signals
,
1975
.