Neural Network based Speech Assistance tool to enhance the fluency of adults who stutter

Millions of adults suffer from a condition called stuttering or stammering. The authors propose the use of a Speech Assistance tool, which helps stuttered speakers achieve higher fluency and a slower rate of speech. The fluency is achieved by adhering to the proposed fluency enhancing technique. The fluency enhancing technique (FET) is inspired by fluency shaping methods and requires the speaker to use a rhythmic method called gentle onset with words and a slower rate of speech. In the training mode, the Speech assistance tool trains an artificial neural network to identify the speaker's FET based words vs. the non-FET or normal words. The audio features are represented using Mel-Frequency Cepstral Coefficients (MFCC), which captures the prosody of the spoken words. In the real-life conversation mode, the speaker gets visual cues to ensure that the speaker adheres to the proposed FET technique. The tool also performs disfluency analysis and provides feedback to users, in terms of FET words ratio, the disfluency score for a hundred words, and the speech rate. The tool also logs the disfluencies periodically to help the speaker track his/her fluency over time. The DTW analysis of MFCC features proven that there is a clear difference in the prosody of the FET and non-FET words. While using the proposed FET based tool, the fluency of the speaker increases and slower speech rate is also achieved. The Speech assistance tool can be used along with Cognitive Behavior Therapy to help rehabilitate adults who stutter.