講演抄録/キーワード |
講演名 |
2019-08-09 10:30
Study on Robust Method for Blindly Estimating Speech Transmission Index using Convolutional Neural Network with Temporal Amplitude Envelope ○Suradej Doungpummet(JAIST)・Jessada Karunjana(NASDA)・Waree Kongprawechnon(SIIT)・Masashi Unoki(JAIST) EA2019-30 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
We have developed a robust scheme for blindly estimating speech transmission index (STI) in noisy reverberant environments based on a convolutional neural network (CNN) with temporal amplitude envelope feature. A method for estimating STI from an observed speech signal is required to predict the speech intelligibility in a sound field where people cannot be excluded. However, there is a significant accuracy reduction of an existing method based on the modulation transfer function due to the mismatch between the models and some real environments. To maintain an appropriate accuracy in general conditions, the robust scheme that the CNN is trained from entire temporal amplitude envelopes of speech signals with multiple noise types and reverberation conditions along with
their associated STIs has been introduced. Simulations were carried out to evaluate the proposed scheme under realistic noisy reverberant conditions. The results showed that the proposed scheme provides high accuracy (i.e., the average root-mean-square error of 0.12 and the correlation of 0.86) under various noise and reverberation conditions.
These results suggest that the proposed scheme can robustly estimate STIs in real noisy reverberant environments. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Speech transmission index / room impulse response / modulation transfer function / temporal amplitude envelope / convolutional neural network / / / |
文献情報 |
信学技報, vol. 119, no. 163, EA2019-30, pp. 47-52, 2019年8月. |
資料番号 |
EA2019-30 |
発行日 |
2019-08-01 (EA) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
EA2019-30 |