WEKO3
アイテム
Minimum generation error training by using original spectrum as reference for log spectral distortion measure
https://nitech.repo.nii.ac.jp/records/3405
https://nitech.repo.nii.ac.jp/records/340584d71d2a-8014-4ba5-871a-0f1065f4a586
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
本文_fulltext (208.9 kB)
|
c2009 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
|
| Item type | 会議発表論文 / Conference Paper(1) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2012-11-07 | |||||||||||
| タイトル | ||||||||||||
| タイトル | Minimum generation error training by using original spectrum as reference for log spectral distortion measure | |||||||||||
| 言語 | en | |||||||||||
| 言語 | ||||||||||||
| 言語 | eng | |||||||||||
| 資源タイプ | ||||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_5794 | |||||||||||
| 資源タイプ | conference paper | |||||||||||
| 著者 |
Wu, Yi-Jian
× Wu, Yi-Jian
|
|||||||||||
| 著者別名 | ||||||||||||
| 識別子Scheme | WEKO | |||||||||||
| 識別子 | 464 | |||||||||||
| 識別子Scheme | NRID | |||||||||||
| 識別子URI | http://rns.nii.ac.jp/nr/1000020217483 | |||||||||||
| 識別子 | 1000020217483 | |||||||||||
| 姓名 | Tokuda, Keiichi | |||||||||||
| 言語 | en | |||||||||||
| 姓名 | 徳田, 恵一 | |||||||||||
| 言語 | ja | |||||||||||
| 姓名 | トクダ, ケイイチ | |||||||||||
| 言語 | ja-Kana | |||||||||||
| 姓 | Tokuda | |||||||||||
| 言語 | en | |||||||||||
| 姓 | 徳田 | |||||||||||
| 言語 | ja | |||||||||||
| 姓 | トクダ | |||||||||||
| 言語 | ja-Kana | |||||||||||
| 名 | Keiichi | |||||||||||
| 言語 | en | |||||||||||
| 名 | 恵一 | |||||||||||
| 言語 | ja | |||||||||||
| 名 | ケイイチ | |||||||||||
| 言語 | ja-Kana | |||||||||||
| 書誌情報 |
en : ICASSP 2009. IEEE International Conference on Acoustics, Speech and Signal Processing, 2009. p. 4013-4016, 発行日 2009 |
|||||||||||
| 出版者 | ||||||||||||
| 出版者 | Institute of Electrical and Electronics Engineers | |||||||||||
| 言語 | en | |||||||||||
| 著者版フラグ | ||||||||||||
| 出版タイプ | VoR | |||||||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||||||
| DOI | ||||||||||||
| 関連タイプ | isIdenticalTo | |||||||||||
| 識別子タイプ | DOI | |||||||||||
| 関連識別子 | http://dx.doi.org/10.1109/ICASSP.2009.4960508 | |||||||||||
| 関連名称 | 10.1109/ICASSP.2009.4960508 | |||||||||||
| 内容記述 | ||||||||||||
| 内容記述タイプ | Other | |||||||||||
| 内容記述 | This paper improves a minimum generation error (MGE) basedHMM training technique for HMM-based speech synthesis by directlyusing the original spectrum instead of line spectral pairs(LSPs) as reference spectrum for log spectral distortion (LSD) measure.Two types of original reference spectra for LSD calculation areinvestigated, including the spectrum extracted from speech waveformby STRAIGHT, and the short-time FFT spectrum calculatedfrom speech waveforms. Since only the harmonics of the FFT spectrumare coincident with the underlying spectral envelope, the LSDbetween generated LSPs and original FFT spectrum is calculated bysampling at the harmonic frequencies, and a weighting function isdesigned to simulate the sampling strategy on LSPs. From the experimentalresults, the MGE-LSD training using the FFT spectrumas reference spectrum achieved the best performance. | |||||||||||
| 言語 | en | |||||||||||
| 内容記述 | ||||||||||||
| 内容記述タイプ | Other | |||||||||||
| 内容記述 | 9-24 April 2009Location: Taipei, Taiwan | |||||||||||
| 言語 | en | |||||||||||
