基于小波变换和支持向量机的音频分类

音频特征提取是音频分类的基础，而音频分类又是基于内容的音频检索的关键。使用小波变换和支持向量机的方法对音频进行分类。研究了小波变换域的音频特征提取，分析了这些特征在小波变换域中的意义。把得到的特征向量作为支持向量机的输入，把音频分成纯语音、带背景音乐的语音、音乐、环境音4种类型。实验结果表明，基于小波域的特征计算简单、能够较好地区分不同的音频类型，得到较高的分类精度。

Audio classification based on wavelet transform and support vector machine

Feature extraction is the foundation of audio classification, while audio classification is a key of content based audio retrieval. Wavelet and support vector machine are used to classify the audio clips. The feature extraction of wavelet domain and the significance of each feature in wavelet domain are analyzed. The feature vectors are used as inputs of support vector machine to classify audio clips into pure speech, speech with music, music and environment sounds. Experimental results show that features extracted form wavelet domain of audio clips are simple to compute and can well separate different type of audio clips, yielding high classification accuracy.