论文部分内容阅读
语音识别技术可以为要求双手同时作业的操作人员和残疾人提供一种便捷的控制方法。作者在文中提出了一种通过结合二阶频率滤波和RASTA技术来增强语音识别鲁棒性的方法,并将这种方法成功应用于机器人化护理床的控制系统中,增强了识别系统在医院、工厂等非稳定噪声环境下语音识别的鲁棒性。通过将HMM/GMM混合模型的传统Mel频率倒谱系数为特征值的识别系统与HMM/GMM混合模型的RASTA-FF2为特征值的识别系统进行比较,并分别在纯语音和带噪语音条件下进行测试,发现经过二阶频率滤波后的FF2特征值再经过RASTA滤波器滤波,特别是在非稳定噪声环境下,以RASTA-FF2为特征值的识别系统比传统的识别系统的识别率更高,这表明FF2特征值与RASTA滤波器技术相结合,一个作用于频域,一个作用于时间域,可以有效地消除语音信号中的不同噪声成份。
Speech recognition technology provides a convenient method of control for operators and people with disabilities who require both hands to work simultaneously. In this paper, the author proposes a method to enhance the robustness of speech recognition by combining the second-order frequency filtering and RASTA technology. The method is successfully applied in the control system of robotic nursing bed and enhances the recognition system in hospitals, Factories and other non-stationary noise environment speech recognition robustness. By comparing the recognition system of eigenvalues based on traditional Mel frequency cepstrum coefficients in HMM / GMM hybrid model with the RASTA-FF2 eigenvalue recognition system of HMM / GMM hybrid model and comparing them with those of pure speech and noisy speech The test results show that the FF2 eigenvalue after second-order frequency filtering is filtered by RASTA filter. Especially in the case of unsteady noise, the recognition system based on RASTA-FF2 has higher recognition rate than the traditional one , Which indicates that the FF2 eigenvalue is combined with the RASTA filter technique, one acting on the frequency domain and one acting on the time domain, which can effectively eliminate different noise components in the speech signal.