Audio Demonstrations
Publication:
Yousefian, N. and Loizou, P. (2012). "A Dual-Microphone Speech Enhancement Algorithm Based on the Coherence Function," IEEE Trans. Audio, Speech, Language Processing, 20(2), 599-609. [pdf]
SNRs are set at the sound sources before convolving the input signals with HRTFs.
Example Sentence 1 (One noise source)
|
Method / Masker at 90 degree |
Speech-shaped noise |
Multi-talker babble |
Female competing talker |
|
Unprocessed Front Mic. (0 dB SNR) |
|||
|
Unprocessed Rear Mic. (0 dB SNR) |
|||
|
Processed via Coherence |
Example Sentence 2 (One noise source)
|
Method / Masker at 135 degree |
Speech-shaped noise |
Multi-talker babble |
Female competing talker |
|
Unprocessed Front Mic. (0 dB SNR) |
|||
|
Unprocessed Rear Mic. (0 dB SNR) |
|||
|
Processed via Coherence |
Example Sentence 3 (Two noise sources)
|
Method / First masker at 90 and the second at 180 degree |
Multi-talker babble and speech-shaped noise |
Speech-shaped noise and female competing talker |
|
Unprocessed Front Mic. (0 dB SNR) |
||
|
Unprocessed Rear Mic. (0 dB SNR) |
||
|
Processed via Coherence |