Signals were saved in MATLAB using wavwrite using 32 bits. According to the MATLAB help file:
"Note that for floating point data where NBITS < 32, amplitude values are clipped to the range -1.0 <= Y < +1.0."
Therefore, in order to avoid clipping of the data, the signals are normalised between +1 and -1, i.e., in MATLAB code: signal = signal./max(signal).

Files:

TimeVarianceSpeech.mp4, TimeVarianceSpeech.mpg:
Q=4 AR parameters were extracted from the utterance ``The farmer's life'' using aryule() in MATLAB. Using roots(), the complex poles corresponding to the parameters were evaluated. The movie illustrates relatively smooth variation of the extracted speech poles and parameters with time. This can be particularly well seen in the pole space for voiced phonemes (/aa/, /axr/, and /ay/).