噪声背景下语音识别特征参数选择研究

为提高语音识别系统的实用性与实时性,该文采用缺失特征分量的方法研究了美尔频率倒谱系数静态特征及其一阶差分各分量对识别率的影响.在不同信噪比情况下,分别对含白噪声、粉红噪声、车载噪声和工厂噪声这四种典型噪声的语音进行了实验研究,结果表明:在保证系统有较高识别率的情况下,在低信噪比时,含白噪声的语音信号缺失美尔频率倒谱系数...

A Study on the Selection of Feature Parameters for Speech Recognition Used in Noisy Environment

To impsore practicability and real-time performance of speech recognition system, the illfluencc of each feature component in Mel frequency cepstral coefficient （MFCC） and first-order differential MFCC （AMFCC） -upon recognition ：rate is studied by＇ using feature component deletion method in this paper. Noisy speech signals under different SNR in four representative noisy environments, such as white noise, pitlk Hoise, Volvo noise and faetolqc：noise are used in the simulations below. The simulation results show that, low older MFCC and AMFCC 1.etlns deleted is availabie in white noisy environment under＂ low $NR for higher recogtlition rate, while.several high order MFCC terms and AMFCC telrns deficiency have rare influence on Lccognitit,rl rate in Volvo noisy environment. But when noisy speech signal is in pink or factory noisy environment, every component is needed.