摘要
音频自动分类是解决音频结构化问题和提取音频内容语义的重要手段之一,是当前基于内容的音频检索领域的一个研究热点。在考察音频数据特征的基础上,针对左-右密度隐马尔可夫模型(left-right DHMM)不能很好反映音频中状态反复的缺点,提出了一种基于各态历经混合高斯密度隐马尔可夫模型(EMGD_HMM)的分类器,并应用于语音、音乐和它们的混合声音的分类。实验结果表明,EMGD_HMM的分类精度要优于left-right DHMM。
Automatic audio classification is one of the significant methods to extract content semantics from audio. An improved classifier based on EMGD_HMM(Ergodic Mixed Gaussian Density Hidden Markov Model) is proposed to classify audio in speech, music, and their mixture. The experimental results show that compared with left-right DHMM(left-right Density Hidden Markov Model), EMGD HMM achieves better classification accuracy.
出处
《电声技术》
2007年第11期52-54,60,共4页
Audio Engineering