Chinese Accent Detection Method Research In Noisy Environment

Accent is an important prosodic feature. The current Chinese accent detection methods can be roughly divided into three categories according to features. One kind detection method is based on the features of acoustic. Another kind detection method is based on grammar features. There is one kind method which fusion acoustic features and grammar features. These three kinds of accent detection method have been widely applied in real life as a key technology of some phonetic system. The application effect is not ideal for the noise in real life.From the influence of the accent detection method according to the features, this paper studies four level research work of Chinese accent detection. There are acoustic features based on context, the child period of joining together short-time spectrum features based on the Perceptual Linear Predictive, the child period of joining together short-time spectrum features based on the Mel Frequency Cepstral Coefficients, and integration of various kinds of optimization featuresChinese accent detection method is based on the acoustic features with the context. Chinese accent intensity is easily affected by the surrounding characters. So this paper selects the eight different context windows for each measured characters. It extracts the pitch, intensity, energy, duration and eight kind context window features.Chinese accent detection method is based on PLP sub-segments splicing short-time spectrum features and MFCC sub-segments short-time spectrum features. Speech frames are divided into several sub-sections on average. Accent detection method divides speech frames into 1 to 20 sub-sections on average based on the PLP and MFCC period of splicing short-time spectrum features. It extract the each period of the maximum, minimum, and average.Chinese accent detection methods fusion optimizing features. This paper introduces a kind of feature selection algorithm. It includes four basic steps: subset produce, subset evaluation, stop condition and result validation. This paper fuses the different optimization features. It includes the fusion of acoustic features and PLP short-time spectrum features, the fusion of acoustic features and MFCC short-time spectrum features, the fusion of MFCC short-time spectrum features and PLP short-time spectrum features, the fusion of acoustic features and short-time spectrum features.In noisy environment, the highest Chinese accent detection accuracy can reach 88.3% in the fusion of acoustic features and short-time spectrum features.