Two-level TDNN (TLDTDNN) technique for large vocabulary Mandarin FINAL recognition

Abstract

A Two-Level Time-Delay Neural Network (TLTDNN) technique has been developed to recognize all Mandarin Finals of the entire Chinese syllables. The first level discriminates the vowel-group based on (a,e,i,o,u,v) and the nasal-group based on nasal ending, (-n,-ng,-others). Orthogonal combination of the two groupings in the first level enables the second level discrimination of all 35 Mandarin Finals. The technique was thoroughly tested with 8 sets of 1265 isolated Hanyu Pinyin syllables, with 6 sets used for training and 2 sets used for testing. The overall result shows that a high recognition rate of 95.3% for inside testing and 93.9% for outside testing is achievable.