Inverse Filtering of Pathologic Voice

Seminar Type: Speech Processing, Seminar (3 SE)
Student: Paul Berghold

Short description

The clinical examination of voice disorders demands for measuring oscillation patterns of the vocal folds. Laryngeal High-Speed Videos (LHSVs) provide the most accurate approach for observing these patterns, but suffer from being invasive: An endoscope is inserted into the mouth of a patient, way back to the pharynx, which makes the examination displeasing. As a consequence, an acoustic method that provides data similar to LHSVs would be desirable. Inverse filtering [1] seems to be a promising approach here. However, the estimation of valid voice source parameters from an acoustic signal is often crucial because models of pathologic voice production are poorly validated on empiric data. State-of-the-art model driven approaches struggle with that fact. Aim of the thesis is to investigate inverse filtering methods and validate estimated voice source parameters with LHSVs.

Your tasks

  • Review of literature
  • Evaluating existing inverse filtering procedures
  • Train methods to fit LHSV data
  • Documentation

Your profile/prerequisites

  • Interest in speech production
  • Speech Communication 1/(2)
  • MATLAB

References

[1] P. Alku, “Glottal inverse filtering analysis of human voice production — A review of estimation and parameterization methods of the glottal excitation and their applications,” vol. 36, no. October, pp. 623–650, 2011.

Contact

Philipp Aichinger (philipp.aichinger@meduniwien.ac.at or 01 40400 1167)