Related Experiment Videos
Decision-directed speech power spectral density matrix estimation for multichannel speech enhancement
Yu Gwang Jin1, Jong Won Shin2, Nam Soo Kim3
1Corporate R&D Center, SK Telecom Co., Ltd., 65 Eulji-ro, Jung-gu, Seoul 04539, Korea ygjin@sk.com.
The Journal of the Acoustical Society of America
|April 5, 2017
Summary
This study introduces a novel multichannel decision-directed method for estimating speech power spectral density (PSD) matrices. This approach enhances noise reduction performance in multichannel speech enhancement systems.
Area of Science:
- Signal Processing
- Acoustics
- Speech Technology
Background:
- Accurate estimation of clean speech power spectral density (PSD) matrices is vital for effective multichannel speech enhancement.
- Conventional methods like maximum likelihood estimators face challenges in tracking dynamic speech characteristics.
Purpose of the Study:
- To propose a multichannel decision-directed approach for estimating the speech power spectral density (PSD) matrix.
- To improve the robustness and performance of multichannel speech enhancement filters.
Main Methods:
- A multichannel decision-directed algorithm is developed to estimate the speech PSD matrix.
- The proposed method contrasts with conventional maximum likelihood estimators.
Main Results:
- The decision-directed method demonstrates robust tracking of time-varying speech characteristics.
- Improved noise reduction performance is observed across diverse noise conditions.
Conclusions:
- The proposed multichannel decision-directed approach offers a more robust and effective solution for speech PSD matrix estimation.
- This method enhances the overall performance of multichannel speech enhancement systems.