| 低信噪比下二值掩蔽算法性能分析 |
| |
| 引用本文： | 蒋毅,梁维谦,周宏,冯振明.低信噪比下二值掩蔽算法性能分析[J].清华大学学报(自然科学版),2012(5):636-641. |
| |
| 作者姓名： | 蒋毅 梁维谦 周宏 冯振明 |
| |
| 作者单位： | 清华大学电子工程系;总后勤部军需装备研究所 |
| |
| 摘 要： | 基于计算听觉场景分析,对基于能量的二值掩蔽语音分离算法的性能进行分析,证明了理想二值掩蔽算法在信噪比下具有最佳的单元分离性能,并通过3种类型带噪语音的分离实验证实了该结论。采用理想二值掩蔽算法对8种噪声类型的低信噪比带噪语音进行了分离实验,信噪比平均提升幅度大于10dB,表明算法对低信噪比语音分离的有效性和普遍适用性;采用非均匀、均匀两种多子带分析滤波器组进行分离性能对比测试,结果表明子带均匀性对信噪比提升影响不大。分析滤波器组的子带数量应大于32以实现较好的分离性能。
|
| 关 键 词： | 语音分离 听觉场景分析 理想二值掩蔽 gammatone滤波器组 |
Performance of binary time-frequency masks in low signal to noise ratio environments |
| |
| JIANG Yi,LIANG Weiqian,ZHOU Hong,FENG Zhenming.Performance of binary time-frequency masks in low signal to noise ratio environments[J].Journal of Tsinghua University(Science and Technology),2012(5):636-641. |
| |
| Authors: | JIANG Yi LIANG Weiqian ZHOU Hong FENG Zhenming |
| |
| Institution: | 1(1.Department of Electronic Engineering,Tsinghua University, Beijing 100084,China; 2.Quartermaster Equipment Research Institute,General Logistics Department,Beijing 100082,China) |
| |
| Abstract: | In the computational auditory scene analysis(CASA) system,the performance of the binary masks algorithm depends on the sound energy which is limited for low signal to noise ratio(SNR) conditions.The ideal binary masks algorithm is shown to have the best SNR performance of all binary masks based on the T-F units.A mixed speech database was set up with eight kinds of noise with SNR of-15,-10,-5 and 0 dB.Speech segregation based the ideal binary masks algorithm improved the average SNR by more than 10 dB indicating very good performance in noisy conditions.The evenness of the filter banks had little effect on the binary masks.The filter banks should have more than 32 channels to improve the segregation ability. |
| |
| Keywords: | speech segregation computational auditory scene analysis(CASA) ideal binary masks gammatone filter banks |
| 本文献已被 CNKI 等数据库收录！ |
|
