Department of Electrical Engineering, The University of Texas at Dallas, Richardson, Texas 75080, USA.
J Acoust Soc Am. 2013 Mar;133(3):1607-14. doi: 10.1121/1.4789891.
A monaural binary time-frequency (T-F) masking technique is proposed for suppressing reverberation. The mask is estimated for each T-F unit by extracting a variance-based feature from the reverberant signal and comparing it against an adaptive threshold. Performance of the estimated binary mask is evaluated in three moderate to relatively high reverberant conditions (T60 = 0.3, 0.6, and 0.8 s) using intelligibility listening tests with cochlear implant users. Results indicate that the proposed T-F masking technique yields significant improvements in intelligibility of reverberant speech even in relatively high reverberant conditions (T60 = 0.8 s). The improvement is hypothesized to result from the recovery of the vowel/consonant boundaries, which are severely smeared in reverberation.
提出了一种单耳二进制时频（T-F）掩蔽技术来抑制混响。通过从混响信号中提取基于方差的特征并将其与自适应阈值进行比较，为每个 T-F 单元估计掩蔽。使用人工耳蜗使用者的可懂度听力测试，在三个中等至相对较高的混响条件（T60=0.3、0.6 和 0.8 s）下评估估计的二进制掩蔽的性能。结果表明，即使在相对较高的混响条件（T60=0.8 s）下，所提出的 T-F 掩蔽技术也能显著提高混响语音的可懂度。这种改进据推测是由于元音/辅音边界的恢复，这些边界在混响中严重模糊。
