A comprehensive survey of automatic dysarthric speech recognition

Abstract

The  need  for  automated  speech  recognition  has  expanded  as  a  result  of significant  industrial  expansion  for  a  variety  of  automation  and  human-machine  interface  applications.  The  speech  impairment  brought  on  by communication  disorders,  neurogenic  speech  disorders,  or  psychological speech  disorders  limits  the  performance  of  different  artificial  intelligence-based systems. The dysarthric condition is a neurogenic speech disease that restricts the capacity of the human voice to articulate. This article presents a comprehensive  survey  of  the  recent  advances  in  the  automatic  dysarthric speech  recognition  (DSR)  using  machine  learning (ML) and  deep  learning(DL)paradigms.   It   focuses   on   the   methodology,   database,   evaluation metrics,and major findings from the study of previous approaches. From the literature  survey  it  provides  the  gaps  between  exiting  work  and  previous work  on  DSR  and  provides  the  future  direction  for  improvement  of  DSR. The  performance  of  the  various  machine  and DLschemes  is  evaluated  for the DSR on UASpeech dataset based on accuracy, precision, recall,and F1-score.  It  is  observed  that  the DLbased  DSR  schems  outperforms  the MLbased DSR schemes.