Mandarin Stops Classification Based on Random Forest Approach

The non-stationary behavior makes stops classification one of worthy examining subject in the speech community. Over several decades, many researchers have sorted out a list of acoustic properties that are useful to identify a stop. In this paper, we extract features that are sufficient to represent the important acoustic properties of stops, like statistic moments of the burst spectrum. In combining a recent developed learning approach, the random forest, we conduct a 6-way classification task to classify Mandarin stops. After a series of bootstrap trials, experimental results demonstrate the superior performance of random forest on the stop classification task over some well-known approaches.