|
An Initial System for
Integrated Synthesis of Mandarin, Min-nan, and
Hakka Speech 國語、閩南語、客家語之整合式語音合成系統 Hung-Yan Gu, e-mail: guhy@mail.ntust.edu.tw 感謝國科會計畫支援, 2006/07 |
A. Examples of Synthetic
Mandarin Speech B. C.
D. E. F.
(國語合成語音例子) (使用固定的合成單元，即409種國語音節中的每一種，都只 錄音、儲存一次) Min-Nan
model adapted to Mandarin 閩南語訓練之模型調適成國語模型 文字檔 合成語音 合成語音 Original
Mandarin-trained model 原始國語語料訓練之模型 文字檔 合成語音 合成語音 Direct concatenation of recorded syllables (原始錄音音節直接串接): zia_ren_de_shou.wav Linear (線性) Piece-wise linear (片段線 性) Direct concatenation of recorded syllables (原始錄音音節直接串 接): dou_ziang_dien.wav
* Fixed
synthesis unit, i.e., each Mandarin syllable is uttered and recorded only
once.
Mandarin has 409 different syllables when tones not
distinguished.
The 409 syllables are uttered by a female.
Pitch-Contour
Model (基 週軌跡模型)
Time-warping
method (時 間軸校正方法)
Synthesis
Method: TIPW (a variant of PSOLA), 合 成方法: TIPW
B.
Examples of Synthetic Min-Nan Speech (閩南語合成語音例子) (1) (2) (3) C. Examples of Synthetic Hakka (sea-land
accent) Speech (1) Using adapted
Min-Nan pitch model (使用調適過的閩南語基週軌跡模型) (2) Using adapted
Min-Nan pitch model (使用調適過的閩南語基週軌跡模型)
* Fixed
synthesis unit.
Syllables are uttered by a female.
(海陸腔客家語合成語音例子)
* Fixed
synthesis unit.
Syllables are uttered by a male.
D. Examples of
Synthetic Hakka (four-country accent) Speech (1) Using adapted
Min-Nan pitch model (使用調適過的閩南語基週軌跡模型) (2) Using adapted
Min-Nan pitch model (使用調適過的閩南語基週軌跡模型) 信號波形 合成 (a)
TIPW variant
of PSOLA 基週軌跡 模型 (a)HMM
& ANN 振幅和音 長 rule-based Synthesis flowchart (合成流程圖):
(四縣腔客家語合成語音例子)
* Fixed
synthesis unit.
Using the same syllables as in C.
signal waveform synthesis
(b) piece-wise linear time warping
pitch-contour model
(b)adapt Min-nan trained model to Hakka
amplitude & duration
