National Language Support Vector Machine-based Language Identification

Speech Recognition is an essential research direction of speech signal processing, and it is also an important part of pattern recognition. As a major aspect of speech recognition, Language Identification(LID) is being paid more attention with the continuous development of speech recognition technology. In order to improve the recognition effect, Language Recognition need to eliminate the individual differences of pronunciation in the same language, and try to find the different acoustic characteristics among the different languages.Using the SVM as a classifier to design a single-feature single-classifier and a multi-feature single-classifier language recognition system. In this experiment, a system identified five languages (Chinese, Naxi, Bai, Miao and Zang) from the databases. First of all, the speech characteristics of the National languages are analyzed to find the differences among various languages. Second, the characteristic coefficients of the speech are picked up and select a different set of characteristic parameters of SVM classifier as input. then, the voice sample characteristic parameters are set, respectively circuit training and testing.We use three single-features and one multi-features as the input features of SVM. In the experiments, including male experiments and female experiments, And we detailed analyze of the experimental results. Experimental results show that acoustic characteristics in the use of different concentration, for a single feature, the pitch frequency in the experiment have the best recognition rate, the reason may be due to pitch frequency reflects information of the excitation source. For the different languages, tone has the largest contribution to the recognition, so it's the highest recognition rate. In addition, the recognition rate of fusion feature was significantly higher than any single feature in this experiment. For language recognition, the reason has many uncertain factors of the language. So far, that is not found a feature to completely identification a different language, so integration of multiple features provides more information on the differences for the classification to reach better results.