Dysarthria Voice Disorder Detection Using Mel Frequency Logarithmic Spectrogram and Deep Convolution Neural Network

Dysarthric speech recognition (DSR), often known as DSR, is an important tool that enables persons with vocal impairments to participate in voice-based automation systems and human-computer interaction. As a result of the poor intelligibility of handicapped speakers, the limited availability of datasets, and the low intra-class and inter-class variability in the speech samples, DSR is an essential component. This research provides a DSR based on the Mel Frequency Logarithmic Spectrogram (MFLS) and Deep Convolutional Neural Network (DCNN). The suggested MFLS+DCNN offers an improved representation of the voice signal in terms of its spectral and temporal characteristics. The results of the MFLS-DCNN scheme are validated using the UASpeech dataset, which was developed based on accuracy, recall, precision, and F1-score. With an accuracy of 96.83%, a recall performance of 0.97, a precision performance of 0.96, and an F1-score of 0.97, the scheme that is proposed has shown a considerable improvement above the conventional state-of-the-art.