Acoustic-phonetic analysis of speech following a lobectomy or esophagectomy

Abstract

An observational study was conducted to monitor and detect the onset of altered level of consciousness and delirium in patients after a major non-cardiac thoracic surgery. Speech recordings from patients were recorded pre-operatively and once daily in the post-operative setting until hospital discharge. A vocabulary consisting of vowels in a coarticulatory neutral environment is used to identify differences in timing or typical articulatory movements that would be indicative of impairment due to delirium. More than one recording session across days is needed in order to compare the patients' speech to their own speech to identify patterns and changes for that talker. Programming has been developed that utilizes the Praat software package to take acoustic-phonetic measurements of pitch, formant frequencies, and amplitude every 6 ms. This processing has generated tens of thousands of rows of data across 1,166 h-vowel-d words reflecting typical patterns for each of the 16 subjects with more than one recording session. This unique dataset has led to insights that go beyond the original intent of identifying the potential onset of delirium including patterns of dyspnea resulting from the surgery. The initial findings from mining the extensive dataset will be presented.