Automatic breath sound detection and removal for cognitive studies of speech and language

Speech has been previously investigated as means of gaining insight into certain psychiatric disorders. Correlation has been found between temporal characteristics of speech and negative symptoms associated with schizophrenia. However the presence of breath sounds in speech may lead to a decreased performance of classification between patient and control groups. This study presents an algorithmic approach for both breath sounds detection and removal, and also analyses its impact on the ability of a Linear Discriminant Analysis (LDA) classifier to discriminate between schizophrenic patients and control subject. Results demonstrate that more accurate feature extraction yielded to a 6.7% increase in discrimination ability from 67.5% to 74.2% to differentiate between schizophrenic patients and control subject based on speech alone.