Bringing advanced speech processing technology to the clinical management of speech disorders

Research to date has made it clear that many speech disorders will not be fully remediated without intensive and long-term intervention. At the same time, practitioners of speech-language pathology grapple with large caseloads and heavy documentation requirements that may leave them unable to provide an adequate duration of treatment for each client. Advanced speech processing technologies have the potential to act as an extender of a clinician’s services by facilitating practice in between treatment sessions, or to enhance the diagnostic accuracy of their evaluations. While there has been a proliferation of apps and software to support speech intervention, few have the capacity for automated or semi-automated assessment of the quality or accuracy of the user’s speech outputs. This special issue highlights current work that applies automatic speech recognition (ASR) or other speech processing technologies to the clinical assessment and/or treatment of disorders affecting speech production. It also discusses current limitations and barriers and identifies potential future directions for technological enhancement of the clinical management of speech disorders.

In the first contribution, McKechnie et al. (2018) provide a systematic review of literature investigating automated analysis tools for the diagnosis and treatment of speech production disorders. While providing an overview of diverse previous studies, their paper also highlights the limited scope of existing work and calls for expanded focus on this topic. In the second paper, Keshet (2018) offers an overview of the architecture of a modern ASR system, aimed at readers without technical expertise in this area. By familiarising the reader with the basic characteristics of ASR and its associated evidence base, these two papers lay a foundation for the advances described in the rest of the issue.

The next two papers explore the possibility of integrating automated speech analysis into mobile apps for use in the treatment of paediatric speech disorders. Ahmed et al. (2018) investigate how children and speech-language pathologists responded to several tablet-based games that had been modified so that gameplay is controlled by the accuracy of the user’s speech. Campbell, Harel, Hitchcock, and McAllister Byun (2017) lay groundwork for an automated extension of an existing app to provide visual biofeedback for the treatment of rhotic misarticulation. Despite the preliminary nature of these studies, both take a positive view of the potential for apps to serve as a source of feedback and motivation and thereby increase treatment doses delivered to children in speech therapy.

The next paper, by Niziolek and Kiran (2018), describes a semi-automated analysis of variability in vowel formants produced by people with aphasia. They use these acoustic findings to make an argument about the nature of speech-motor control in this population. Similarly, Buz, Buchwald, and Keshet (2018) argue that measurements of voice onset time (VOT) can provide insight into the level at which speech production is disrupted in patients with aphasia and/or apraxia of speech, and they report a procedure to automatically extract VOT measurements from large samples of either typical or clinical speech.

The final two papers focus on the population of individuals with dysarthria secondary to amyotrophic lateral sclerosis (ALS). As elicitation of extended speech samples is fatiguing and thus undesirable in the context of progressive speech disorders, it would be beneficial if automated analyses can enhance the diagnostic accuracy of brief samples of speech. Wang et al. (2018) argue that a single acoustic and articulatory sample can be used to predict the severity of speech impairment in individuals with ALS. Likewise, Rong, Yunusova, Richburg, and Green (2018) offer evidence that articulatory data from an alternating motion rate (AMR) task can be used to identify early signs of speech-motor degeneration in this population. Jointly, these papers show that automatically extracted acoustic and articulatory features can be leveraged to improve the process of diagnosing and monitoring speech deficits.

These contributions, along with similar recent research, point to a promising future in which technologies for automated speech processing can improve the diagnosis and treatment of speech disorders. At the same time, this line of inquiry is in a very early stage, and our contributors are uniform in calling for increased interdisciplinary collaboration to advance this important field. Keeping pace with advances in speech recognition technology will allow the discipline of speech-language pathology to stay relevant while also helping to secure improved quality of life for patients with communication deficits.

TARA MCALLISTER
Department of Communicative Sciences and Disorders, New York University, New York, NY, USA
[email protected]

KIRRIE J. BALLARD
Faculty of Health Sciences, The University of Sydney, Sydney, Australia