【摘要】 考虑到变异语音产生的非线性特点,本文提出了一种基于TEO能量算子倒谱特征的应力影响下变异语音识别方法。先将语音信号分割成21个不同频带的信号,然后计算TEO能量,最后进行对数运算和离散余弦变换。对航空模拟飞行器中采集的小词表特定人的识别实验,采用非线性分析的基于TEO能量算子倒谱特征的方法,能有效地提高变异语音的识别性能,比传统的基于MFCC特征的方法识别率提高了11.3％。更多还原

【Abstract】 On the basis of nonlinear feature of the stressed speech, an approach of TEO based cepstrum coefficients (TEOCEP) is proposed for recognition of speech under G-Force, in which the speech signal is first divided into 21 subbands, and then the TEO energies are estimated. Finally, the TEOCEP feature is obtained by using log-compression and discrete cosine transform (DCT) on the estimated TEO energies. For recognition experiments of speaker dependent, small vocabulary and stressed speech collected in an aero-flight simulator, the proposed method gets an improvement of 11.3% over the method of using traditional linear MFCC feature.更多还原