Speech-enabled environmental control in an AAL setting for people with speech disorders: a case study

The homeService research project is concerned with developing personalised speech-enabled interfaces, in an AAL setting, for users with severe physical impairments and associated disordered speech. By putting state-of-the-art speech recognition systems into people's homes, invaluable lessons can be learned from doing long-term trials `in-the-wild'. The experiences gained from the first homeService user's case story is described here. Each system is initially deployed with acoustic models adapted using a relatively small amount of enrolment data. During use, data is subsequently collected as the user interacts with the system and this data is used to update the models at a later stage. This paper contrasts results from experiments carried out online with the live system, and offline with the collected data. Particular emphasis is put on the amount of adaptation data as well as the use of manual vs. automatic annotations in the context of trying to ensure that the implementation and personalisation strategy will scale with many users.