Time Series Similarity Measurement And Its Application In Speech Recognition

Time series is a set of data points arranged in chronological order,each element consists of acquisition time and data value(Where the data value can be a certain number or a vector,that is,multivariate.),is a common data type in production and life.It is widely used in mathematical statistics,signal processing,pattern recognition and other application science and engineering related to time data measurement.Similarity measurement can be used to guide production practice.In this paper,the most common time series in daily life----speech sequence data is taken as an object to study the similarity measurement of time series.For the unary speech signal similarity measurement and multivariate speech signal similarity measurement methods were improved,an improved algorithm for recognizing speech signals based on time series similarity was proposed.The main work of this paper is as follows:(1)Aiming at the similarity measurement method of unary speech signals,this paper firstly performs polynomial smoothing filter(Savitzky-Golay)on speech signals in the pre-processing stage,and then selects important points for the smoothen sequences.In this paper,based on the use of continuous three-point value algorithm,a new extreme point selection constraint is added to the selection of important points,and an obvious turning point is introduced to effectively reduce the length of time series.Experiments show that the proposed algorithm is feasible and effective.(2)In view of the similarity measurement method of multiple speech signals,In view of the similarity measurement method of multiple speech signals,the traditional endpoint detection algorithm has the problem that it is not good to distinguish the voiceless area from the voiceless area in the low SNR environment.In this paper,the short-time zero-crossing rate algorithm in endpoint detection is improved to make the distinction between the voiceless area and the voiceless area more obvious.In the speech signal recognition,the distance calculation of each frame between sequences is introduced into the cosine distance calculation,which improves the problem that the single-dimensional value change in the traditional Euclidean distance affects the experimental results.The experimental results show that the improved method proposed in this paper is not only more accurate,but also more efficient in recognition time.