The LetsRead Corpus of Portuguese Children Reading Aloud for Performance Evaluation

The LetsRead Corpus of Portuguese Children Reading Aloud for Performance Evaluation Jorge Proença1,2, Dirce Celorico1, Sara Candeias3, Carla Lopes1,4, Fernando Perdigão1,2 1 Instituto de Telecomunicações, Coimbra, Portugal 2 Department of Electrical and Computer Engineering, University of Coimbra, Portugal 3 Microsoft Language Development Centre, Lisbon, Portugal 4 Polytechninc Institute of Leiria, Leiria, Portugal IT, Department of Electrical and Computer Engineering, University of Coimbra - Pole II, 3030-290, Coimbra, Portugal E-mail: {jproenca,dircelorico}@co.it.pt, t-sacand@microsoft.com, {calopes ,fp}@co.it.pt Abstract This paper introduces the LetsRead Corpus of European Portuguese read speech from 6 to 10 years old children. The motivation for the creation of this corpus stems from the inexistence of databases with recordings of reading tasks of Portuguese children with different performance levels and including all the common reading aloud disfluencies. It is also essential to develop techniques to fulfill the main objective of the LetsRead project: to automatically evaluate the reading performance of children through the analysis of reading tasks. The collected data amounts to 20 hours of speech from 284 children from private and public Portuguese schools, with each child carrying out two tasks: reading sentences and reading a list of pseudowords, both with varying levels of difficulty throughout the school grades. In this paper, the design of the reading tasks presented to children is described, as well as the collection procedure. Manually annotated data is analyzed according to disfluencies and reading performance. The considered word difficulty parameter is also confirmed to be suitable for the pseudoword reading tasks. Keywords: Children’s speech, Reading disfluencies, European Portuguese 1. Introduction To evaluate the reading aloud performance of children, human assessment is usually involved, where a teacher or tutor has to take time to individually determine the performance in terms of fluency (speed, accuracy and expression). The automatic estimation of reading ability can be an important alternative or complement to the usual methods, and can improve other applications such as e- learning. However, there are no computer assisted applications for European Portuguese (EP) that automatically evaluate the reading aloud performance of children. Techniques must be developed to analyze audio recordings of read utterances by children and detect the deviations to the intended correct reading (disfluencies). Even for other languages, this automatic evaluation is a developing topic, and the focus is on reading of isolated words instead of larger sentences (Black et al., 2011; Duchateau et al., 2007). To carry out the goals of the LetsRead project1, we found it necessary to create a large new speech corpus of EP children’s speech with utterances of reading tasks that are rich in the common disfluencies that children commit while reading. There are some children's speech databases for EP, such as Speecon with rich sentences (Speecon Consortium, 2005); ChildCAST (Lopes, 2012; Lopes et al., 2012) with picture naming; the Contents for Next Generation (CNG) Corpus targeting interactive games (Hämäläinen et al., 2013) and (Santos, 2014; Santos et al., 2014) with child- adult interaction. However, these databases do not present the required samples of disfluent reading speech. Since children's speech …

Content truncated to 500 words