Pathological speech deformation degree assessment based on integrating feature and neural network

In tasks related to the analysis and recognition of pathological speech it is often more important to provide the respective person (e.g. physician) with guidelines for a deformation degree assessment of speech signal than to achieve a very accurate automated recognition. By ear it is easy to judge whether the speech is regular or deformed, but any attempt of a deformation degree evaluation is not satisfactory. According to above status, we presented a deformation degree assessment system of speech signal based on integrating feature and neural network. The system is comprised of four main sections, a pre-processing section, a feature extracting section, a neural network processing section and assessment value calculation section. And also this paper integrates different speech features to calculate the perceptual distance vector to improve assessment ratio, the perceptual distance between the pathological speech and the normal speech under test is used as input to the neural network. The simulation results demonstrate that a classification accuracy of 95% is obtained with database of 80 speech signals (40 normal and 40 pathological cases).