摘要
针对单个语音去噪算法在去噪过程中关注点较为单一,而多个语音去噪算法在融合时存在细节信息被削弱、融合效果不理想的问题,提出一种多个语音去噪算法下的自适应门限融合策略,将带噪信号分别经过3种不同的去噪算法得到3个去噪信号;根据自适应门限值以帧为单位进行帧筛选,得到自适应门限融合策略下的去噪信号;为提高识别效果,采用倒谱提升器对Gammatone频率倒谱系数(Gammatone Frequency Cepstrum Coefficient, GFCC)进行改进,并联合支持向量机进行噪声环境下的语音识别。实验结果表明,在5、10、15、20 dB四种信噪比下,通过该融合策略所得到的去噪信号与目前主流的顺序融合及多级融合方式相比,在语音识别率方面平均提高3.6%,融合倒谱提升器的GFCC特征相比于GFCC特征平均提高了2.2%。
To solve the problem that a single speech denoising algorithm has a single focus in the process of denoising,while the details of multiple speech denoising algorithms are weakened and the fusion effect is not ideal,an adaptive threshold fusion strategy under multiple speech denoising algorithms is proposed.Three denoising signals are obtained by using three different denoising algorithms.Frame filtering is carried out according to the adaptive threshold,and then the denoising signal under the adaptive threshold fusion strategy is obtained.On the other hand,in order to improve the recognition effect,the Gammatone Frequency Cepstrum Coefficient(GFCC)is improved by using a cepstrum lifter,and the speech recognition in noisy environment is carried out jointly with support vector machine.The experimental results show that under the four signal to noise ratios of 5,10,15 and 20 dB,the speech recognition rate of the denoised signals obtained by the fusion strategy is improved by 3.6%on average compared with the current mainstream sequential fusion and multistage fusion methods,and the GFCC feature using cepstrum lifter is improved by 2.2%on average compared with the GFCC feature.
作者
薛珮芸
师晨康
白静
赵建星
汪思斌
XUE Peiyun;SHI Chenkang;BAI Jing;ZHAO Jianxing;WANG Sibin(College of Electronic Information and Optical Engineering,Taiyuan University of Technology,Jinzhong 030600,China;Postdoctor Workstation,Shanxi Institute of Advanced Innovation,Taiyuan 030002,China)
出处
《无线电工程》
2024年第4期1026-1033,共8页
Radio Engineering
基金
山西省应用基础研究计划项目(201901D111094)
山西省基础研究计划(20210302124544)。
