Computerised objective measurement of strain in voiced speech

Abstract

Voice quality assessment is required by healthcare professionals in patients suffering from voice problems. Speech and language therapists (SLTs) use a well-known subjective assessment approach which is called GRBAS, to quantify voice problems. GRBAS is an acronym for a five dimensional scale of measurements of voice properties which were originally recommended by the Japanese Society of Logopeadics and Phoniatrics and the European Research for clinical and research use. The properties are `Grade', `Roughness', `Breathiness', `Asthenia' and `Strain'. In requiring the services of trained SLTs, this subjective assessment make the GRBAS measurement expensive to administer. In this research, computerised objective measurement of `Strain' in voice using two regression prediction models is compared with measurements produced by SLTs according to the GRBAS scale. These regression models are K Nearest Neighbor Regression (KNNR) and Multiple Linear Regression (MLR). These new approaches for prediction of Strain are based on different subsets of features, different sets of data, and different prediction models in comparison with previous approaches in the literature. The best feature subset for predicting Strain objectively was obtained amongst different feature subsets. When compared with the mean of five SLT's scores, over 102 samples, the computerised measurement was found to have a Normalized Root Mean Square Error (NRMSE) averaged over 20 trials, lower than that of each individual SLT. We have achieved a NRMSE of 14.6% and 15.1% for the MLR and KNNR respectively when the best feature subsets were used for predicting Strain objectively.

Machine Learning Applied to GRBAS Voice Quality Assessment

“…The required filtering may be achieved either by digital filters or an FFT. A high value of LH with cut-off frequency 1.5 kHz can be indicative of asthenia [36] and strain [37] due to imperfectly functioning vocal cords damping the spectral energy of formants above 1.5 kHz. LH measurements with a cut-off frequency of 4.0 kHz are useful for detecting breathiness and voicing since the spectral energy of voiced speech (vowels) is mostly below 4.0 kHz.…”

Acoustic analysis and digital signal processing for the assessment of voice quality

“…• Strain is the perceived effect of a person speaking, or trying to speak with abnormality functioning vocal cords [26]. This is probably the most subjective GRBAS component with largely variable effects.…”