Glottal source estimation: Methods of applying the LF-model to inverse filtering

The incorporation of the LF model into closed-phase inverse filtering based estimation of glottal source waveforms is considered. The LF model is a five-parameter model representing the effective voice source over open and closed glottal phases. Three techniques for LF model fitting are described: two based on Prony´s method and one using gradient descent. The estimation methods are evaluated using glottal flow waveforms obtained synthetically and through inverse filtering. Under appropriate conditions, all of the estimation methods produce good fits of the LF model to the inverse filtered waveform. In noise, gradient descent produces more consistent fits than the more tractable Prony-based techniques.<>