ISCA Archive ISCSLP 1998
ISCA Archive ISCSLP 1998

Character Error Correction for Chinese Speech Recognition System

Liqin Shen, Haixin Chai, Yong Qin, Donald Tang

With continuous speech recognition technology, one can input documents to computer very fast. Unfortunately, speech recognition can not achieve 100% accuracy, and users have to correct the recognition errors. Therefore the average throughput, which is the number of correct characters inputted to computer per minute taking the time for error correction into consideration also, drops dramatically. The smaller error rate and the better error correction mechanism, the less time will be required for error correction. Here we will focus on the second factor, the better error correction mechanism, to improve the throughput.