國立虎尾科技大學 |

Low Complexity Spectral Imputation for Noise Robust Speech Recognition.

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Low Complexity Spectral Imputation for Noise Robust Speech Recognition./
作者:	van Hout, Julien.
面頁冊數:	75 p.
附註:	Source: Masters Abstracts International, Volume: 50-06, page: .
Contained By:	Masters Abstracts International50-06.
標題:	Engineering, Electronics and Electrical. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=1510374
ISBN:	9781267333926

Low Complexity Spectral Imputation for Noise Robust Speech Recognition.
van Hout, Julien.

Low Complexity Spectral Imputation for Noise Robust Speech Recognition. - 75 p.

Source: Masters Abstracts International, Volume: 50-06, page: .

Thesis (M.S.)--University of California, Los Angeles, 2012.

With the recent push of Automatic Speech Recognition (ASR) capabilities to mobile devices, the user's voice is now recorded in environments with a potentially high level of background noise. To reduce the sensitivity of ASR performance to these distortions, techniques have been proposed that preprocess the speech waveforms to remove noise effects while preserving discriminative speech information. At the expense of increased complexity, recent algorithms have significantly improved recognition accuracy but remain far from human performance in highly noisy environments.

ISBN: 9781267333926Subjects--Topical Terms:

845382
Engineering, Electronics and Electrical.

Low Complexity Spectral Imputation for Noise Robust Speech Recognition.
LDR:03203nam 2200337 4500 001 713084
005 20121003100412.5
008 121101s2012 ||||||||||||||||| ||eng d
020 $a 9781267333926
035 $a (UMI)AAI1510374
035 $a AAI1510374
040 $a UMI $c UMI
100 1 $a van Hout, Julien. $3 845720
245 1 0 $a Low Complexity Spectral Imputation for Noise Robust Speech Recognition.
300 $a 75 p.
500 $a Source: Masters Abstracts International, Volume: 50-06, page: .
500 $a Adviser: Abeer Alwan.
502 $a Thesis (M.S.)--University of California, Los Angeles, 2012.
520 $a With the recent push of Automatic Speech Recognition (ASR) capabilities to mobile devices, the user's voice is now recorded in environments with a potentially high level of background noise. To reduce the sensitivity of ASR performance to these distortions, techniques have been proposed that preprocess the speech waveforms to remove noise effects while preserving discriminative speech information. At the expense of increased complexity, recent algorithms have significantly improved recognition accuracy but remain far from human performance in highly noisy environments.
520 $a With a concern for both complexity and performance, this thesis investigated ways to reduce the corruptive effect of noise by directly weighting the power-spectrum (SMFpow) or log-spectrum (SMFlog ) of speech by a mask whose values are within [0,1] and are indexed on the local relative prominence of speech and noise energy. Additional contributions include a low-complexity approach to mask estimation and the use of spectral flooring for matching the dynamic range of clean and noisy spectra. These two techniques are evaluated on two standard noisy ASR databases: the Aurora-2 connected digits recognition task with 11 words, and the Aurora-4 continuous speech recognition task with 5000 words.
520 $a On the Aurora-2 task, the SMFlog algorithm leads to state-of-the-art performance, with a limited complexity compared to existing techniques. The SMFpow technique, however, results in many insertions that we attribute to the rather weak language model present in the Aurora-2 setup. On the Aurora-4 task, both algorithms show significant improvements over the un-enhanced baselines. In particular, word-accuracies obtained with SMFpow approach those of a state-of-the-art front-end algorithm, on half of the noise types. Yet, the performances are heavily noise dependent, suggesting that the proposed technique is effective only given a good initial mask estimation.
520 $a This study confirms the potential of techniques that are based on direct spectrum masking, and proposes a framework for doing so. Future work will need to consider more elaborate mask estimation techniques to further improve on the performance.
590 $a School code: 0031.
650 4 $a Engineering, Electronics and Electrical. $3 845382
650 4 $a Computer Science. $3 593922
690 $a 0544
690 $a 0984
710 2 $a University of California, Los Angeles. $b Electrical Engineering 0303. $3 845570
773 0 $t Masters Abstracts International $g 50-06.
790 1 0 $a Alwan, Abeer, $e advisor
790 1 0 $a Pottie, Gregory $e committee member
790 1 0 $a Wortman Vaughan, Jennifer $e committee member
790 $a 0031
791 $a M.S.
792 $a 2012
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=1510374