Research Of Speech Identity Recognition Based On Synergetics Algorithm

With the widespread use of computer applications, man-machine interaction based on speech is more and more needed, but there are many factors like noise in the speech adversely affecting its effect of processing. Improving performance of speech processing and recognition rate is the core of the research for interaction technology based on speech.On the basis of analyzing the existing speaker-recognition technology based on speech, we designed a scheme of speech recognition based on synergetics with the introduction of blind signal-separation and synergetic method. The profound research on the critical modules of this scheme is shown in the following aspects:(1). After analyzing the existing speaker-recognition technology based on speech, we designed a speech recognition scheme based on synergetics and constructed an overview model to cover its shortcomings. The model consists of speech preprocess, feature data extraction, feature data interpretation and identity recognition.(2). Blind signal separation was inducted into the existing speech preprocess. We designed and realized a set of algorithms for signal de-nosing, sampling, quantification, pre-amplifying, framing. endpoint detection, etc.(3). As for feature data extraction, we designed and tested an integrated method using Linear Prediction Cepstrum Coefficient and Mel-Frequency Cepstrum Coefficient in improvement of the current way.(4). When interpreting the feature data, synergetics theory and K-means clustering algorithm were employed. A data-interpretation model based on synergetics was established. We designed and examined the key components of the model like prototype vector selection based on K-means clustering algorithm, and feature data interpretation based on synergetics.(5). In doing identity recognition, interpretation based on synergetics was used in acquiring value of speech feature. Then the feature value was compared to the filed specifics of a speaker to judge their identity information. We designed and conducted the identity judging on the basis of similarity.(6). All the aspects from sampling to recognition performance were analyzed, and we did simulation experiment on the de-noising capability and robustness of the scheme. It was approved that our way of speech recognition was more accurate and efficient than the existing methods, especially in the noisy environment and after changing of speech.