论文部分内容阅读
用于音乐和语音的识别方法不适用于非结构化环境声音事件的识别。提出一种基于隐马尔可夫模型(HMM)和支持向量机(SVM)的二层分类策略,对家庭保健监测系统中的语音、警报声、电话铃声、笑声、尖叫声和咳嗽声等6种声音事件进行识别。首先,提取Mel频率倒谱系数(MFCCs)来分析环境声音信号。其次,以提取的MFCC特征为输入,依次采用HMM和SVM构造二级分类模型,通过识别和确认两个过程来对家庭保健监测系统中的环境声音事件进行识别。实验证明,该技术能提高家庭保健监测系统中易混淆环境声音的识别率。
Recognition methods for music and speech are not suitable for the identification of acoustic events in unstructured environments. This paper proposes a two-level classification strategy based on Hidden Markov Model (HMM) and Support Vector Machine (SVM) to detect voice, alarm, telephone ring, laughter, scream and cough in home health monitoring system 6 kinds of sound events to identify. First, extract Mel Frequency Cepstral Coefficients (MFCCs) to analyze environmental sound signals. Secondly, taking the extracted MFCC features as input, the HMM and SVM are used to construct the second-level classification model in turn, and the environmental sound events in the home health monitoring system are identified by identifying and confirming the two processes. Experiments show that this technology can improve the recognition rate of ambiguous environmental sounds in home health monitoring system.