Development of vanilla LSTM based stuttered speech recognition system using bald eagle search algorithm

Stuttering or stammering is considered as the most important parameter in the speech recognition algorithm. For the conversion of stuttered speech into readable text, first it is vital to detect the stuttered speech. In many existing models, the exactness of the recognition system is degraded because of the additional noise present in the speech signals. Therefore, proposed a bald eagle search algorithm based vanilla long-short term memory (BES-vanilla LSTM), a system for identifying the stuttered speech among the number of speech signals. Initially, the dataset collected from the TORGO database undergoes preprocessing phase for the elimination of unwanted noise signals present in the input speech signals using the spectral subtraction method. Further, the preprocessed signals are passed to the feature extraction phase. In this phase, the pitch frequencies are extracted from the input signals. Then the extracted pitch frequencies are passed to the input layer of the Vanilla LSTM for stuttered speech recognition. Using the fitness value of the BES algorithm, the LSTM recognizes the stuttered speech and is given at the output layer. The proposed system is implemented in the python tool. The comparison of the suggested system's performance is made with the other existing system by metrics such as f1-score, recall, accuracy and precision, and the overall efficiency of the designed system is studied.