Study of locus equations as features for speech classification and recognition
Reid, Clarice Virginia
MetadataShow full item record
Classification of speech is a difficult problem due to the continuous and variable nature of speech. A Locus Equation is a linear regression model that relates F2 at the start of a CV vowel transition to F2 in the middle of the vowel, where C is held constant and V is varied to cover the vowel space. The resulting equation, which takes the form y = mx + b, contains information about the consonant and the consonant transitions. In this thesis, Locus Equations are examined as a potential feature for speech classification and regression problems. The equations are automatically generated, and then used as features in classification of dialects and speech dysarthria. Although the equations did not perform well with the dialect classification, results were promising for classification of speakers with dysarthria.