A comparative study of intelligent voice quality assessment using impedance and acoustic signals

Objective assessment techniques for classifying voice quality for patients recovering from treatment for cancer of the larynx should lead to more effective recovery than the present approach, which is very subjective and depends heavily on the experience of the individual Speech and Language Therapist (SALT). This work follows an earlier study where an Artificial Neural Network (ANN) was trained on parameters derived from electrolaryngograph electrical impedance (EGG) signals recorded while a patient was phonating /i/ as steadily as possible, and gave an indication of voice quality inline with the standard UK Speech and Language Therapist (SALT) seven point scale. The applicability of this approach to voice quality assessment of acoustic signals is described, and the results are found to compare very well with those derived from the impedance signals. It was also noted that for both the impedance and the acoustic signals, the ANNs were able to classify the very good (recovered) and the very poor (abnormal) voices well, but performed quite badly with the mid-range classifications, raising questions about the accuracy of these classifications.