Combining prosodic and spectral features for Mandarin intonation recognition

In this paper, a feature set for Chinese Mandarin intonation is addressed. We first analyze the prosodic features, which make significant contributions to distinguishing between Chinese Mandarin interrogative intonation and declarative intonation. Then make a fusion between prosodic features and Mel Frequency Cestrum Coefficient (MFCC) feature. Our study is performed on a Mandarin question corpus, which contains large amount and various types of interrogative sentences. Experimental results indicate that the combination of prosodic features and MFCC feature achieves 77.09% classification accuracy, a 16% improvement over that of prosodic features alone. The proposed feature set performs well in the condition of speaker-independent.