一种基于人耳听觉感知和子带补偿滤波的鲁棒语言辨识特征参数提取算法

针对目前语言辨识系统所采用的特征参数没有充分考虑人耳听觉机制、鲁棒性较差的问题,提出一种符合人耳听觉感知特性的鲁棒语言辨识参数提取算法.该算法主要从两个方面提高特征参数的鲁棒性:在计算各子带能量时采用更符合人耳感知特性的Gammachirp滤波器组代替常用的三角滤波器组;为每一子带通道设计一个补偿滤波器.子带补偿滤波器的设计采用数据驱动的策略,通过补偿使得各子带滤波器输出信号的失真及环境噪音导致的失真同时达到最小.实验表明,文中所提出的特征在常见噪声环境下,性能均优于目前普遍使用的Mel频率倒谱系数特征及其衍生参数.

A Robust Feature Parameter Extraction Algorithm for Language Identification Based on Audio Perception and Sub-Band Compensation Filtering

In current language identification system,the commonly used feature parameters have not made the best use of auditory characteristics and have weak robustness in complex environments.An auditory-based robust feature extraction algorithm is proposed.Each sub-band energy of the extracted auditory features is calculated by using a Gammachirp filter bank instead of the commonly used triangle filter bank.The compensation filter using data-driven analysis for each sub-band output is obtained by a constrained optimization process which jointly minimizes the environmental distortion as well as the distortion caused by the filter itself.Experimental results show that the feature outperforms the Mel-frequency cepstral coefficient widely used in noisy environments.