Recognition of Score Word in Freestyle Kayaking

Speech is the most natural information carrier for human beings, and it is likely to become the main way of human-computer interaction in the future. This paper presents an isolated score word recognition method using Mel-scale Frequency Cepstral Coefficients (MFCC) and Dynamic Time Warping (DTW). The processing stage of the speech signal is the basic stage of the speech recognition system, to analyze the speech signal and convert it into speech feature parameters. An endpoints detection method is proposed using the joint adjustment of short-term energy and zero-crossing rate. It can better detect the endpoints, and directly improve the accuracy of subsequent work. On this basis, the MFCC feature is then extracted from the preprocessed speech signal, and the DTW pattern matching is applied to the extracted features. In the experiments, speeches from multiple speakers were collected, each with a specific freestyle kayak action word. The results show that this method has better performance comparing with the existing methods.