A Framework for Collecting Realistic Recordings of Dysarthric Speech - the homeService Corpus

A Framework for Collecting Realistic Recordings of Dysarthric Speech - the homeService Corpus Mauro Nicolao1, Heidi Christensen1, Stuart Cunningham2, Phil Green1, and Thomas Hain1 1Computer Science; University of Sheffield, United Kingdom 2Human Communication Sciences, University of Sheffield, United Kingdom {m.nicolao, h.christensen, s.cunningham, p.green, t.hain}@sheffield.ac.uk Abstract This paper introduces a new British English speech database, named the homeService corpus, which has been gathered as part of the homeService project. This project aims to help users with speech and motor disabilities to operate their home appliances using voice commands. The audio recorded during such interactions consists of realistic data of speakers with severe dysarthria. The majority of the homeService corpus is recorded in real home environments where voice control is often the normal means by which users interact with their devices. The collection of the corpus is motivated by the shortage of realistic dysarthric speech corpora available to the scientific community. Along with the details on how the data is organised and how it can be accessed, a brief description of the framework used to make the recordings is provided. Finally, the performance of the homeService automatic recogniser for dysarthric speech trained with single-speaker data from the corpus is provided as an initial baseline. Access to the homeService corpus is provided through the dedicated web page at http://mini.dcs.shef.ac.uk/resources/homeservice-corpus/. This will also have the most updated description of the data. At the time of writing the collection process is still ongoing. Keywords: dysarthric speech corpus, speech disorders, speech corpus collection, realistic environment recordings, automatic dysarthric speech recogniser. 1. Introduction The world of digital devices is becoming more and more oriented towards speech-enabled interfaces. Speech is an attractive alternative to more traditional input devices, such as remote controls, keyboards, or PC mice especially when environmental or physical constraints make conventional interfaces difficult to use. The success of speech-driven digital devices often depends on the recognition accuracy of the interface. Good results have been obtained for speakers with typical speech, how- ever speech recognition can be challenging for people who have disordered speech associated with neuro-motor condi- tions. These people are also less likely to be able to use con- ventional interfaces due to their impaired motor function. Performance is adversely affected by the increased vari- ability that characterises disordered and dysarthric speech. Hence it is essential to have access to representative audio signals with which the acoustic model, at the core of state- of-the-art automatic speech recognition (ASR) techniques, can be trained or adapted (Sharma and Hasegawa-Johnson, 2012; Christensen et al., 2012). Very few databases exist, and one cannot assume that other speakers – typical or disordered – will provide a good enough match to the speech of a particular target user. The speech of some individuals is simply so atypical that it is hard to find good matches in terms of acoustic similarity to include into a good baseline model (Christensen et al., 2014). Most of the existing dysarthric speech databases, such as the TORGO (Rudzicz et al., 2011) and UAspeech (Kim et …

Content truncated to 500 words