This paper presents a new speech recognition algorithm using template matching. The system uses Feature Space Trajectory Distance (FSTD) and semantic processing as the basis of templates matching. Rule based semantic ...
详细信息
This paper presents a new speech recognition algorithm using template matching. The system uses Feature Space Trajectory Distance (FSTD) and semantic processing as the basis of templates matching. Rule based semantic abstraction in traffic area is the key to distinguish the relation of the words to be recognized in our system, instead of HMM statistical method. SVM classifier is used as categorizing algorithm. Considering the limitation of general recognitions on vocabulary templates that need tens of thousands of word templates, we use syllable template and scalable merging algorithm instead, only need about 1400 syllables. The algorithm has been applied in our scalable syllable interpolation-merging algorithm, and gets a high accuracy up to 99.43% in the simulation experiment with former three results. We also add our lab's semantic address parser to strengthen the system's ability of mining address names. The technology has been accepted and it will act an important role in traffic query system in Peking 2008' Olympics
暂无评论