【摘要】 针对现有婴儿哭声识别方法在噪声环境下和不同婴儿间鲁棒性不足的问题,提出一种婴儿哭声识别的优化算法。利用子带谱熵法端点确定婴儿哭声信号的有效区间,增强算法在噪声环境下的鲁棒性;从有效哭声信号中提取平滑Mel频率倒谱系数(SMFCC)作为特征参数;基于SMFCC构建婴儿哭声模板,增强算法在不同婴儿间的鲁棒性;使用动态时间规整算法(DTW)计算哭声信号与模板的距离,得到识别结果。实验表明:在具有噪声的婴儿哭声测试集中,哭声识别准确率均达到72%以上,该方案在噪声环境下和不同婴儿个体的哭声识别中表现出较强的鲁棒性。更多还原

【Abstract】 In order to overcome the shortcomings that the current infant crying recognition algorithms are not robust enough in a noisy environment and among different infants,an optimization algorithm for infant crying recognition is proposed.This method uses the subband spectral entropy method to detect the effective range of infant crying to improve the robustness of algorithm in noisy backgrounds.The smooth Mel frequency cepstrum coefficient( SMFCC) is extracted from the effective crying signals as the characteristic parameters among different infants.The infant crying template is constructed based on the SMFCC parameters to improve the robustness of the algorithm.The dynamic time warping algorithm( DTW) is used to calculate the distance between the crying signal and the template and obtain the recognition result.Experiments show that in the test set in noisy background,the crying recognition accuracy rate is more than 72%,the method shows strong robustness in the noise environment and different infants crying recognition.更多还原