Automatic Glottal Inverse Filtering of Speech (Larnyx, Voice).

The flow of air through the glottis, the glottal volume-velocity, reflects the action of the vocal folds and is thus an important indicator of laryngeal function. However, the glottal volume-velocity waveform cannot be measured directly because of the filtering effects of the vocal tract. A method to estimate the glottal volume-velocity (v-v) waveform during human speech by automatic glottal inverse filtering has been developed. The method uses a sampled electroglottograph (EGG) signal to identify regions of glottal closed phase in a simultaneously sampled speech pressure signal. The EGG has been shown by this study to provide a more robust means to identify the closed phase than methods based on analysis of the speech waveform. A least squares covariance analysis of the closed phase speech segments is used to determine the parameters of the digital inverse filter necessary to remove the effect of the vocal tract from the speech waveform. The problem of low frequency noise is reduced by use of a high pass filtering electret microphone, followed by low frequency compensation by a digital inverse filter that models the microphone's frequency response. The filter parameters are determined by a truncated singular value decomposition (SVD) solution of a linear backward prediction over the closed phase of the uncompensated glottal v-v waveform. The remaining low frequency drift is removed by a cubic spline interpolation fit to points of known closed phase on the glottal v-v waveform. This method of compensating a high pass microphone is shown to reduce the effect of low frequency breath and ambient noise even in speech samples recorded in a noisy environment. The experimental results have shown that this method provides a reliable estimate of the glottal v-v waveform for normal and abnormal speech. This automatic and non-invasive method provides an easily obtained measure that can be an important tool for the speech and laryngeal pathologist. It is believed that, with time, the method will make an important contribution to the study of laryngeal pathology and aid in its early detection.PhDBiomedical engineeringUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/160437/1/8502946.pd