论文部分内容阅读
利用听觉感知机理，建立一个基于听觉感知机理的语音信号特征提取模型。本文由两部分组成，一部分是在传统听觉计算模型基础上提出听觉倒谱特征AFCC（AnditoryFrequencyCepstralCoefficient）的提取方法，这样既压缩了特征维数，减小计算量，又使各个特征维之间相互独立，满足HMM模型的要求。并且根据听觉神经中枢的长时整合特性，文中提出了用低通滤波模型来模拟这种功能。结合该低通模型，提取的语音信号的听觉倒谱特征在HMM框架下取得较好的鲁律性。另一部分在研究听觉侧抑制机理的基础上，提出一个简单有效的听觉侧抑制处理模型。美尔倒谱特征MFCC谱特征经过该侧抑制模型处理，得到侧抑制美倒谱特征MFCCI，实验表明，该新特征MFCCI鲁棒性能比MFCC有大大提高。听觉倒谱特征AFCC经过该侧抑制处理得到侧抑制听觉倒谱特征AFCCI，实验表明，该新特征AFCCI鲁律性能比AFCC有大大提高。
Using auditory perception mechanism, a speech signal feature extraction model based on auditory perception mechanism is established. This paper consists of two parts, one is based on the traditional auditory computational model proposed auditory cepstrum feature AFCC (AnditoryFrequencyCepstralCoefficient) extraction method, which not only reduces the number of features to reduce the amount of computation, but also to make each feature independent of each other , Meet the requirements of HMM model. According to long-term integration of the auditory nerve center, a low-pass filter model is proposed to simulate this function. Combined with the low-pass model, the auditory cepstrum feature of the extracted speech signal achieves better robustness under the HMM framework. On the other hand, based on the study of auditory side inhibition mechanism, a simple and effective auditory side suppression model is proposed. After the MFCC spectral features of the cepstrum feature are processed by the side suppression model, the MFCCI of the side-suppressed cepstrum feature is obtained. Experiments show that the MFCCI robustness of the new feature is greatly improved compared with the MFCC. The auditory cepstrum feature AFCC obtained the side-suppressed auditory cepstrum feature AFCCI through this side-suppression treatment. Experiments show that the new feature AFCCI Lu Ru-performance significantly improved than AFCC.