计算机应用研究
計算機應用研究
계산궤응용연구
APPLICATION RESEARCH OF COMPUTERS
2014年
11期
3316-3319,3323
,共5页
可调Q-因子小波变换%语音增强%清浊音分离%Donoho阈值%阈值函数
可調Q-因子小波變換%語音增彊%清濁音分離%Donoho閾值%閾值函數
가조Q-인자소파변환%어음증강%청탁음분리%Donoho역치%역치함수
tunable Q-factor wavelet transform(TQWT)%speech enhancement%separation of voiced signal and unvoiced sig-nal%Donoho threshold%threshold function
针对语音增强算法中传统的小波阈值法的局限性,提出一种基于可调Q-因子小波变换和清浊音分离的语音增强算法。首先用过零率和短时能量法判别清音和浊音;然后在可调Q-因子小波变换下,对清、浊音采用不同的阈值处理,在不同尺度上,分别结合系数能量和噪声方差得到的阈值作为清音和浊音的阈值确定准则;再利用改进的阈值函数分别处理清音和浊音的小波系数,估计出不含噪声的系数;最后进行小波逆变换,得到抑制了噪声的语音信号。对含有高斯白噪声和有色噪声的语音进行仿真实验,结果表明:与目前许多经典的去噪方法相比,该方法在去噪效果和提高语音可懂度方面均有一定的改善。
針對語音增彊算法中傳統的小波閾值法的跼限性,提齣一種基于可調Q-因子小波變換和清濁音分離的語音增彊算法。首先用過零率和短時能量法判彆清音和濁音;然後在可調Q-因子小波變換下,對清、濁音採用不同的閾值處理,在不同呎度上,分彆結閤繫數能量和譟聲方差得到的閾值作為清音和濁音的閾值確定準則;再利用改進的閾值函數分彆處理清音和濁音的小波繫數,估計齣不含譟聲的繫數;最後進行小波逆變換,得到抑製瞭譟聲的語音信號。對含有高斯白譟聲和有色譟聲的語音進行倣真實驗,結果錶明:與目前許多經典的去譟方法相比,該方法在去譟效果和提高語音可懂度方麵均有一定的改善。
침대어음증강산법중전통적소파역치법적국한성,제출일충기우가조Q-인자소파변환화청탁음분리적어음증강산법。수선용과령솔화단시능량법판별청음화탁음;연후재가조Q-인자소파변환하,대청、탁음채용불동적역치처리,재불동척도상,분별결합계수능량화조성방차득도적역치작위청음화탁음적역치학정준칙;재이용개진적역치함수분별처리청음화탁음적소파계수,고계출불함조성적계수;최후진행소파역변환,득도억제료조성적어음신호。대함유고사백조성화유색조성적어음진행방진실험,결과표명:여목전허다경전적거조방법상비,해방법재거조효과화제고어음가동도방면균유일정적개선。
Aiming at the limitations of methods on speech enhancement by traditional threshold methods in wavelet domain, this paper proposed a new speech enhancement algorithm based on the tunable Q-factor wavelet transform and separation of voiced signal and unvoiced signal.Firstly,it separated voiced signal and unvoiced signal with zero-crossing ratio and short-time energy.The adaptive threshold values combined the energy of coefficients and the variance of noise in different scales,re-spectively.Then it applied the improved Donoho threshold value and threshold function to process wavelet coefficients of voiced signal and unvoiced signal,and estimated the original coefficients from noisy coefficients.Lastly,it used the inverse transform to obtain the original speech signal which the noise was removed.Comparing with the other current classical algo-rithms,experimental results show that the modified algorithm improves the effect of de-noising and speech intelligibility in white Gaussian noise and colored noise background.