Optimization of stammering in speech recognition applications

In the field of speech recognition systems, existing work focuses only on the classification of speech into a stammering speech or a normal speech. Further processing is only one of the normal, use-able, speech. The goal of this research is to amend stammered speech and enable its general use. It is observed, that when a person stammers, there is a decrease in the amplitude of the person's voice, which can be used to eliminate the repetitions, elongations, and silent intervals. This allows the production of a better speech recognition system. Substantial improvements included in this research compared to previous implementations is coming up with a new deep-learning algorithm, which enhances speech recognition for people suffering from stammering. This is interfaced with a real-time web application. The procedure adopted is to first register the audio, remove stammers, and convert the processed data into recognized speech. Stammering is removed by using an amplitude threshold, which is obtained from a neural network model. The recognized speech is then prepared for further real-world usage.