AN EMPIRICAL STUDY OF MULTIPASS DECODING FOR VIETNAMESE LVCSR
Khoa Trinh, Ha-Thanh Nguyen, Duc Anh Duong, Quan Hai Vu · 2008
ABSTRACT In this paper, we represent an empirical study of multipass decoding for Vietnamese LVCSR. We report our experiments with N-best, lattice and consensus decoding on the VNBN data. Results from this study indicate that our acoustic model for Vietnamese was precise. The results could be investigated in further steps to improve the performance of our system. Index Terms Vietnamese, Acoustic Model, Language Model, N-best, Word Lattice, Confusion Network. 1. INTRODUCTION Large vocabulary continuous speech recognition (LVCSR) is both a pattern recognition and search problem. As described in [4], the acoustic and language models are built upon a statistical pattern recognition framework. In speech recognition, making a search decision is also referred to as decoding. The decoding process should take into account all available knowledge sources when hypothesising an utterance. Besides the speech signal, and the models of the recognition units, also knowledge about syntax, semantic, and other properties of the natural language might be used when searching for the most likely word sequence. One way to include these knowledge sources in the search process is to use them simultaneously to constrain a single search. Since many of the natural language knowledge sources contain long-distance effects, the search can become quite complex. Furthermore, the common left-to-right search strategy requires that also all knowledge sources are formulated in a predictive, left-to-right manner, which restricts the type of knowledge that can be used. One way to solve these problems is to apply sources not simultaneously but sequentially so that the search for the most likely hypothesis is constrained progressively. Thus the advantages provided by a knowledge source can be traded-off against the costs of applying it. First, the most powerful and cheapest knowledge sources are applied to generate a list of the top