Lithuanian Digits Recognition by Using Hybrid Approach by Combining Lithuanian Google Recognizer and Some Foreign Language Recognizers
Tomas Rasymas, Vytautas Rudžionis · Communications in computer and information science · 2015
In this paper we are presenting our results obtained by experimenting with different classification methods which are suitable for creating hybrid speech recognizer. We tried to create Lithuanian digits recognizer which is capable of producing more than 95 % accuracy. Classification methods that were used are: k-Nearest neighbors (5, 11, 15, and 21), Linear Discriminant Analysis, Quadratic Discriminant Analysis, Logistic Regression, Naïve Bayes, Support Vectors classifier. Experiments were taken using five foreign recognizers: English, Russian, two German and Google recognizer for Lithuanian language. Best results were received when Naïve Bayes classifier was used – 97.51 %. Average accuracy of single recognizers: English – 84.9 %, Russian – 80.6 %, German I – 63 %, German II – 81.4 % and Google recognizer – 82.6 %. By using hybrid approach accuracy was increased by 12.61 % compared with best single recognizer result.