Next: Parameter estimation
Up: Linear Prediction analysis
Previous: Linear Prediction analysis
The transfer function of a lossless tube can be described by an all pole
model. This is also a reasonable approximation to speech formed by the
excitation of the vocal tract by glottal pulses (although the glottal
pulses are not spectrally flat).
Figure 37: The lossless tube model of speech production
Nevertheless, with sufficient parameters the LP model can make a
reasonable approximation to the spectral envelope for all speech sounds.
The vocal tract is not built of cylinders
The vocal tract is not lossless
The vocal tract has a side passage (the nasal cavity)
fricatives (e.g. /s/ and /sh/) are generated near the lips
Speech Vision Robotics group/Tony Robinson