基于矢量量化的婴儿哭声识别算法

VQ-based recognition algorithm for babies cries

基于嵌入式微处理器的智能婴儿摇篮的功能需求,本文提出了一种针对婴儿哭声的语音识别算法。语音特征参数使用目前最广泛采用的MFCC参数,使用最大欧氏距离划分初始样本,使用LBG算法不断迭代优化得到最终码本。在婴儿哭声的识别阶段,提取待识别语音的MFCC特征参数与已有码本计算矢量量化误差,若矢量量化误差两次低于判断值,输出判别为婴儿哭声的结果。

For the requirements of embedded microprocessor-based smart baby cradle, In this paper, we bring forward a baby crying for speech recognition algorithms. Speech feature parameters using the most widely used MFCC parameters, using the maximum Euclidean distance the initial sample, using the LBG algorithm iteratively optimize the final code book. In the identification stage of Baby cries , the extraction to be recognized voice of the MFCC feature parameters with the existing code vector quantization error of this calculation, if the vector quantization error to judge the value of two lower than the output determine the outcome for the baby crying.