Current speech recognition services are not suitable for people with speech disorders, which present difficulties in coordinating muscles and articulating words and sentences. In this case, a speaker-dependent approach is strongly required in order to address the specific vocal disarticulation. Several Deep learning approaches have been proposed in the literature to address this problem. However, they require many voice samples of people to properly work, and this is not practical. In this paper, we present an innovative Automatic Speech Recognition (ASR) system which is able to correct failures of deep learning based solution adopting Natural Language Processing (NLP) techniques. The proposed solution can perform both single word and whole sentence corrections by analyzing the speech context. We evaluated the solution in a home automation case study and proved the good accuracy of our model.

A NLP-based Approach to Improve Speech Recognition Services for People with Speech Disorders

Fazio M.
Secondo
Writing – Review & Editing
;
Carnevale L.
Penultimo
Writing – Original Draft Preparation
;
2022-01-01

Abstract

Current speech recognition services are not suitable for people with speech disorders, which present difficulties in coordinating muscles and articulating words and sentences. In this case, a speaker-dependent approach is strongly required in order to address the specific vocal disarticulation. Several Deep learning approaches have been proposed in the literature to address this problem. However, they require many voice samples of people to properly work, and this is not practical. In this paper, we present an innovative Automatic Speech Recognition (ASR) system which is able to correct failures of deep learning based solution adopting Natural Language Processing (NLP) techniques. The proposed solution can perform both single word and whole sentence corrections by analyzing the speech context. We evaluated the solution in a home automation case study and proved the good accuracy of our model.
2022
978-1-6654-9792-3
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11570/3250078
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? ND
social impact