Automatic Speech Recognition System

747 Words2 Pages

Nowadays, many applications for learning using ASR system. ASR can capture children's interest and engaged them in their learning (Husniza. H, Fauziah. A.R., Sobihatun. N.A.S., 2012). ASR also can increase the quality of learning and teaching help make sure e-learning is accessible to all through the cost-effective production of synchronized and captioned multimedia ( Mike. W., 2002). So, IMELDA is one of application using ASR technologies to help challenging young children in primary school in Malaysia. IMELDA was encouraging children to interest in learning English. However, during testing IMELDA at school for six children from primary school some limitation be found in this ASR system that affect accuracy. In order to ensuring this application using ASR technology work properly, it need be able to handle any of the possible caused affect performance an ASR engine (Victoria. Y., 2012). The theory of the literature reviewed relationship between ASR recognition performance as measured by the accuracy where % of the word correct, divided by the total number of words used. ACCURACY OF ASR The accuracy ASR is a challenging problem to handle for Automatic speech recognition system. This difficulty causes from some factor . According Victoria Y (2012) these error are caused by two factors: external and internal factors. An external factor which is noise environment and internal factor from error of components and language model (LM) by ASR system. In IMELDA factor causes ASR problem from both (e.g. Child's voice, noise environment, pronunciation error and language model (LM) used in IMELDA not suitable for L2. Noise environment The noise in the environment is one of the external factors that determine th... ... middle of paper ... ...ion error happens when younger children may not have a correct pronunciation. Furthermore, the model an ASR engine in IMELDA not suitable for children L2 because the phonetic from children L2 is different with children L1. Sometimes young children not known how to articulate specific phonemes (Schotz, 2001). Almost speech recognition system develops using English language. Thus, ASR system gives it problem for children L2 especially in accuracy because style pronunciation English language is different compare children L1. According to Muhirwe. J (2005) to build speech engine we need a corpus. Speech corpus can get from collections of text and speech and both are used basis of the statistic processing of natural language processing (NLP). Cole et al., (1994) also state to develop a speech corpus may involve data collection and transcription.

Open Document