Vietnamese sentence recognition algorithm in embedded device based on specialized transition network

Abstract

In this work, a proposed high speed continuous speech recognition algorithm which was designed towards embedded devices and experimented on Vietnamese will be presented. To be more specific, this algorithm has used a transition network (TN) as search-space, which integrates many language model systems and condensed by the algorithm format to both reduce processing time and memory usage while matching. The final results which were evaluated on 100 speech samples have achieved the high accuracy of 92.24% on an embedded device named WandBoard Rev C1 kit.