多媒体网络语音音调数据特征智能识别方法

Intelligent recognition method for multimedia network voice and tone data features

利用传统方法对多媒体网络语音音调数据特征进行识别,存在识别准确性低,识别速度慢的问题。针对上述问题,提出一种新的多媒体网络语音音调数据特征智能识别方法。该方法分为三部分:第一对输入的多媒体网络语音音调数据进行预处理,包括数据转换、预加重、分帧加窗和端点检测等4步;第二提取预处理数据特征参数,包括基音频率、共振峰、mel倒谱系数;第三利用GMM-SVM模型在提取到的特征参数基础上进行智能识别。结果表明:利用本方法对多媒体网络语音音调数据特征进行智能识别,与基于人工神经网络的语音特征识别方法和基于蚁群算法特征选择的语音识别方法相比,平均正确识别率提高8.3%和12.4%,平均所用时间减少0.7 s和1.24 s,识别速度加快。

The traditional methods for recognizing the features of multi-media network speech tone data have the problems of low recognition accuracy and slow recognition speed.In view of the above problems,a new intelligent recognition method for voice and tone data features in multimedia networks is proposed.The method is divided into three parts:first,the input multimedia network voice tone data is pre-processed,including data conversion,pre-emphasis,sub-frame windowing and endpoint detection steps;second,the pre-processed data feature parameters,including pitch frequency,formant,Mel cepstrum coefficient;third,the GMM-SVM model is used to extract the mentioned parameters.Intelligent recognition is done on the basis of characteristic parameters.The results show that this method can be used to recognize the tone data of multimedia network intelligently.Compared with the speech recognition method based on artificial neural network and the speech recognition method based on ant colony algorithm,the average correct recognition rate is increased by 8.3%and 12.4%,and the average recognition time is reduced by 0.7 s and 1.24 s the recognition speed accelerected.