论文部分内容阅读
为了解决已有语音活动检测方法在噪声条件下性能急剧下降的问题,提出了一种基于稀疏编码的语音活动检测方法.该方法在训练阶段为语音和每种可能的噪声训练字典;在识别阶段首先识别环境噪声类型,然后将语音字典和对应环境噪声字典拼接形成一个大字典用于稀疏分解,最后用语音字典上的稀疏表示实现语音和非语音的判断.该方法一方面引入了噪声类型识别,可以有针对性地选择噪声字典;另一方面引入噪声开集识别机制,能识别新的噪声类型并且为之训练模型.实验结果表明所提方法比传统方法具有更强的噪声鲁棒性.
In order to solve the problem of the speech activity detection method dropping rapidly under noisy conditions, a speech activity detection method based on sparse coding is proposed, which can train the dictionary for speech and every possible noise during the training phase. During the recognition phase Firstly, the type of environmental noise is identified, and then a dictionary is constructed by splicing the speech dictionary with the corresponding ambient noise dictionary for sparse decomposition. Finally, the sparse representation on the speech dictionary is used to determine the speech and non-speech.On the one hand, this method introduces the noise type identification , The noise dictionary can be selected in a targeted way, and on the other hand, an open noise recognition mechanism is introduced to recognize and train the new noise types.The experimental results show that the proposed method is more robust to noise than traditional methods.