Automatic Recognition of Prolongations and Repetitions in Stuttered Speech Using ANN and HMM

Abstract

Stuttering in speech is a speech disorder where normal flow of Speech is disrupted by occurrences of dysfluencies, such as repetitions, interjections, prolongations and so on. Stuttered speech is affecting millions of people in their day to day life. Automatic recognition of speech stuttering technology is a great aid to admit the challenges and is prominent in speech language pathology. This paper describes new approach to automate the stuttering speech recognition with hybrid HMM/ANN technology. Recognition systems based on Hidden Markov Models (HMM) are effective with strong statistical foundation but have a large number of unstructured parameters as they rely on prior assumptions. Artificial Neural Networks (ANN) is used to overcome the limitations of HMM. The current work was carried out on UCLASS database of 20 people with stuttering. Instances of prolongations and repetitions in stuttered speech are recognized by using ANN and HMM. The experimental investigation of resulting hybrid system was successful in identifying the instances of prolongations and repetitions in stuttering speech.