| Summary: | 碩士 === 慈濟大學 === 醫學資訊學系碩士班 === 103 === The use of computers and mobile phones have been popular in the recent decade. The speech recognition function also has been integrated in the computers and mobile phones to assist use. Popular speech recognition systems such as Apple Siri, Android Google voice typing, Microsoft Windows 7 voice recognition, and IBM Viavoice, etc. can help to input general terms easily and the correctness is also acceptable. But for the proper noun, the recognition ability of existing systems is not enough.
In this thesis, a special speech recognition system for proper noun, using in nursing records, is studied. We use the Julius system, a open source large vocabulary CSR engine developed by Kawahara lab. et.al., to develop the speech recognition system. First the HTK (Hidden Markov Model Toolkit) been used to build the acoustic model and then SRILM (Stanford Research Institute Language Modeling Toolkit) been used to establish N-Gram language mode. Finally, we integrated them by Julius system.
Although the correct rate of the recognition is still not acceptable but the system may help to build similar system and to reduce the workload of nurses.
|
|---|
