SIAS: Speech Intelligibility of Atypical Speech
Obtaining speech intelligibility ratings generally involve subjective and time-consuming human annotations. The SIAS studies the possibilities of automatic objective information sources. By doing so, insights as to what speech intelligibility is, can be gained.
Speech intelligibility (SI) is gradually becoming more important in speech therapy and language learning, for evaluating pathological and non-native speech, respectively. However, SI is a complex construct, there are no standard procedures to obtain SI ratings, and current procedures generally involve human annotations, which are subjective and time-consuming.
About the project
In the current project, we will study the possible contribution of automatic, objective information sources: ‘Automatic Speech Recognition’ (ASR) and acoustic phonetic features (APFs). We will study the relations between APFs, human and ASR-based SI ratings for all words together, and for classes of words based on ‘part of speech’ (POS) tags. These APFs will not only provide novel scientific insights into SI, they will also contribute useful information for diagnosis and therapy (for pathological speech) and for language education (for non-native speech).
The general goal of the research is to study to what extent automatic, objective measures obtained by means of automatic speech recognition (ASR) and forced alignment (FA) can be used to acquire novel insights and obtain information about speech intelligibility (SI). For this purpose, the following CLARIAH components are used: KALDI Speech recogniser for Dutch (ASR), Forced Alignment (FoAl) and Alpino & Frog (Parsing).