An Effective Speech Understanding Method with a Multiple Speech Recognizer based on Output Selection using Edit Distance

An Effective Speech Understanding Method with a Multiple Speech Recognizer based on Output Selection using Edit Distance* Kazutaka Shimada a, Satomi Horiguchi a, and Tsutomu Endo a a Department of Artificial Intelligence, Kyushu Institute of Technology 680-4 Iizuka Fukuoka 820-8502, Japan {shimada, s_horiguchi, endo}@pluto.ai.kyutech.ac.jp Abstract. In this paper, we propose a simple and effective method for speech understanding. The method incorporates some speech recognizers. We use two recognizers, a large vocabulary continuous speech recognizer and a domain-specific speech recognizer. The integrated recognizer is a robust and flexible method for speech understanding. For the integration process, we use a simple edit distance measure of each output sentence from each recognizer. Our method has high scalability and accuracy. The experimental results show the effectiveness of the proposed method. Keywords: Multiple speech recognizer, Integration, Output selection, Edit distance. 1. Introduction Speech understanding and dialogue systems have been developed for practical use recently. These systems often recognize user utterances incorrectly. It is important to deal with speech recognition errors for speech understanding systems. Extracting keywords and understanding an utterance using them reduce speech recognition errors (Bouwman et al. 1999, Komatani and Kawahara 2000). Another approach is to use domain-specific grammars and linguistic models. However these methods can not handle out of domain and spontaneous utterances. One approach for the improvement is to repair recognition errors by users. There are many studies on detection of recognition errors in a speech output. Goto et al. (2005) have proposed some systems with nonverbal speech information, such as “SPEECH STARTER” and “SPEECH SPOTTER”. Ogata and Goto (2005) have proposed a speech input interface with a speech-repair function. Although repairing recognition errors by humans is effective in terms of development of a speech understanding system with high recognition accuracy, it is costly for users. Combining some recognizers is one of the best approaches to improve the accuracy of speech understanding systems (Isobe et al. 2007, Utsuro et al. 2004). Utsuro et al. (2004) have obtained high accuracy by using some speech recognizers' outputs. However they dealt with word error reduction only. Although Isobe et al. (2007) have proposed a multi-domain speech recognition system by using some domain-specific recognizers, their system cannot treat out-of-domain utterances such as a chat between users. However the chat utterances often include a significant role as the context of the dialogue. In this paper we propose a simple and effective speech understanding method based on a large vocabulary continuous speech recognizer (LVCSR) and some domain-specific speech * Copyright 2008 by Kazutaka Shimada, Satomi Horiguchi, and Tsutomu Endo. This research was supported by the New Energy and Industrial Technology Development Organization (NEDO), Intelligent RT Software Project, 2008. 350 22nd Pacific Asia Conference on Language, Information and Computation, pages 350–357 recognizers (DSSR). The task of this system is speech understanding for a livelihood support robot. The DSSRs recognize particular utterances about orders; e.g., order utterances from elders who need care and order utterances from nurses. Figure 1 shows the outline of the proposed method. We construct the …

Content truncated to 500 words