Article
Analyzing Impaired Speech in Context of Magnetic Resonance-guided Focused Ultrasound Using Convolutional Neural Networks (en)
* Presenting author
Abstract:
In context of treating neurological diseases such as Parkinson´s Disease and Essential Tremor Magnetic Resonance-guided Focused Ultrasound (MRgFUS) treatment is a relatively new method for reducing the effects of tremor. To investigate potential side effects of this neurosurgical procedure on speech, pre- and post-operative recordings of 56 patients were evaluated. The audio files were subjectively evaluated by means of a survey with 21 participants. The perceived speech quality was rated in five categories: monotony, clearness, fluidity, listening effort and voice tremor. Additionally, a Convolutional Neural Network (CNN) was trained to perform an objectified assessment based on these categories. Therefore, a computer-based objective analysis was utilized to extract features from the audio signals and combine them before optimizing the CNN architecture. The results show small positive and negative changes in speech quality, which were overall balanced and therefore more indicative of natural speech variations related to the patients' underlying diseases. The developed CNN was able to perform the speech quality evaluation using a combination of Mel Frequency Cepstral Coefficients (MFCCs), fundamental frequency, and loudness features, resulting in a Mean Absolute Error (MAE) of 0.81 on a scale from 0 to 10, compared to a MAE of 0.67 for the survey participants.