Recognition of score words in freestyle kayaking using improved DTW matching

Voice is the most natural information carrier for human beings, and is likely to become the main method of human–computer interaction in the future. This article focuses on the recognition of score words in freestyle kayaking, and collects words from multiple speakers, each with a specific freestyle kayak action word. In this paper, a new method using mel-scale frequency cepstral coefficients (MFCC) and improved dynamic time warping (DTW) is presented for isolated speech recognition. An endpoint detection method is proposed and implemented based on short-time energy and zero-crossing rate. After preprocessing with endpoint detection, the speech signal was analyzed and converted into speech feature parameters using MFCC. During the training phase, the signals of the training part were trained, and the labeled features were generated. During the identification phase, we improved the DTW algorithm by using multiple constraints to make path matching within the constraints more accurate. Experiments were conducted and the results showed a high recognition rate for a specific score word in freestyle kayaking. In addition, this method provides relatively good results in noisy environments with high signal-to-noise ratios.