Probabilistic AMC with channel compensation for GSC-based speech enhancementGSC 기반 음성개선을 위한 확률적 AMC 및 채널 보상

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 577
  • Download : 0
The performance of speech interfaces tend to severely degraded by interfering noises. The generalized sidelobe canceller (GSC), one of the microphone array-based noise reduction algorithms, is widely used for non-stationary noise reduction. The GSC consists of the fixed beamforming, the blocking matrix (BM), and the multichannel noise canceller (NC). The BM and the NC filter coefficients should be alternatively updated in order not to cancel the target speech signals from the desired direction. The adaptation mode controller (AMC) decides the adaptation of the NC filter coefficients only for noise intervals where the target speech signals do not exist. The GSC performance largely depends on the AMC accuracy because target speech signals would be distorted by the incorrect adaptation. Thus, in this dissertation, we try to improve the performance of the AMC in the GSC. In the proposed algorithm, a probabilistic AMC is designed and adopted to the GSC. To detect target speech intervals, the proposed AMC calculates the inter-channel normalized cross correlation and estimates the speech absence probability (SAP). Based on the estimated SAP, the adaptation mode of the NC filter is decided. The other reason of the performance degradation is the channel mismatch including the path differences from the target source to the microphone array. This mismatch would decrease the signal-to-noise ratio (SNR) of the target speech signal after beamforming. That is mainly why we propose an efficient channel mismatch compensation method to improve the performance of the beamforming-based noise reduction algorithm. In the proposed method, the channel mismatch is compensated through two steps. At the first step, the fixed channel compensator, for which the transfer function is pre-estimated in a noiseless condition, compensates for the overall channel mismatch. At the second, the time-varying channel mismatch is compensated in real-time by the adaptive channel compensator somewhat pre...
Advisors
Hahn, Min-Sooresearcher한민수researcher
Description
한국과학기술원 : 정보통신공학과,
Publisher
한국과학기술원
Issue Date
2010
Identifier
455457/325007  / 020055187
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 정보통신공학과, 2010.08, [ xii, 108 p. ]

Keywords

Beamforming; Noise reduction; Speech enhancement; GSC; 음성인식; GSC; 빔포밍; 잡음제거; Speech recognition

URI
http://hdl.handle.net/10203/39869
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=455457&flag=dissertation
Appears in Collection
ICE-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0