摘要
在信噪比依赖的非均匀谱压缩(SNSC)鲁棒语音特征提取技术和VTS算法的基础上,该文提出了一种新的MC-SNSC模型补偿算法。SNSC技术是一种根据人类听觉对声音强度-响度感知转化关系的谱幅度变化操作和噪声抑制技术。基于对数谱域的噪声以及SNSC特征提取对语音信号特征所产生的失配函数,推导出了MC-SNSC模型补偿算法。实验证明使用这一新算法,识别率比当前较理想的VTS和PMC算法有很明显的提升,算法的复杂度较VTS等算法仅有轻微的增加。
A novel model compensation method is proposed, which integrates the Vector Taylor Series (VTS) approach with a robust feature extraction technique called SNR-dependent Non-uniform Spectral Compression (SNSC). The SNSC method is a spectral operation of magnitude transformation which resembles the human intensity-to-loudness conversion process and de-emphasizes noisy bands. Based on this mismatch function, which models the effect of the noise onto the clean speech in the Log-spectral domain together with the SNSC, a new model compensation procedure is derived. By adopting this novel model compensation approach, significant improvement over the PMC and VTS method can be found in different additive noisy environments at the expense of slight increase in computational complexity.
出处
《电子与信息学报》
EI
CSCD
北大核心
2007年第6期1384-1388,共5页
Journal of Electronics & Information Technology
基金
国家自然科学基金(60101002
60172048)资助课题
关键词
语音识别
模型补偿
非均匀谱压缩
Speech recognition
Model compensation
Non-uniform spectral compression