Paper Abstract and Keywords |
Presentation |
2011-01-27 15:45
Relation between musical noise generation in nonlinear signal processing and speech recognition performance Ryoichi Miyazaki, Takayuki Inoue, Nobuhisa Hirata, Hiroshi Saruwatari, Kiyohiro Shikano (NAIST), Tomoya Takatani (TOYOTA) SP2010-106 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
In this paper, we discuss a relation between musical noise generation in nonlinear signal processing and
speech recognition performance. Recently, applications using speech recognition, e.g., spoken dialogue robots, have
increased. In past studies, we have commonly used a cepstral distortion as a measure to predict the recognition
performance. However, to measure the cepstral distortion, we need clean speech signals before being added with
noise, but they can not be measured in real environments. Therefore, in this paper, we introduce a kurtosis ra-
tio which is measurable in real environments and corresponding to the musical noise generalization instead of the
cepstral distortion. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Speech recognition / Kurtosis ratio / Blind source separation / Musical noise / Nonlinear noise reduction / / / |
Reference Info. |
IEICE Tech. Rep., vol. 110, no. 401, SP2010-106, pp. 19-24, Jan. 2011. |
Paper # |
SP2010-106 |
Date of Issue |
2011-01-20 (SP) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
SP2010-106 |
|