頁籤選單縮合
題名 | 臺語基本音節合成單元資料庫的發展與建立=Taiwanese Syllable-based Synthesis Units Database |
---|---|
作者 | 佘永吉; 鍾高基; 吳宗憲; | 書刊名 | 中華醫學工程學刊 |
卷期 | 19:1 1999.03[民88.03] |
頁次 | 頁47-58 |
分類號 | 312.23 |
關鍵詞 | 臺語; 合成單元資料庫; 音韻學與數位訊號處理; 語音合成; Taiwanese language; Synthesis units database; Phonology and DSP; Text-to-Speech; |
語文 | 中文(Chinese) |
中文摘要 | 聽語為人類最主要的溝通方式。臺語源自於閩南語,全世界將近有五千萬人口能 使用閩南語,超過 75% 的臺灣人口能使用臺語。 雖然學者提出多種臺語文字化的方案,但 長期欠缺科學化及資訊化的專研,嚴重的影響臺灣文化的發展、母語教育之推動、乃至溝通 輔助科技的開發。根據語言學、音韻學以及數位訊號處理,本研究的目的:發展以單音節為 基礎的臺語語音合成單元資料庫及語音合成系統。臺語為聲調語言,音節由母音、子音與聲 調所構成。本研究首先探討適用之基本音素表,分析可能的音節組合與對應的漢字語意單元 ,以建立單聲調合成單元資料庫。合成單元音檔的建立,係透過載字句的錄音而載字句的篩 選:資料庫中的所有合成單元,未必具有語意,語音訊號的音節界線必須清晰且利於切割。 錄音過程中,速度、音高、音長與音量盡量維持其一致性。經由過零率與能量曲線切割,建 立 522 個合成單元資料庫。 文字轉語音系統由文字輸入、字轉音規則庫、合成單元電子音 檔及語音輸出等子系統所構成;輸入漢字或臺語現代文,經由字轉音規則對應語音資料庫, 然後輸出臺語語音。初步評估的結果顯示:單聲調資料庫大致包含所有具語意的臺語音節, 此資料庫實現單聲調文字轉語音系統。本研究提供臺語語音分析及合成的基礎資料庫,並可 應用於母語教育與溝通輔助稍技的發展。 |
英文摘要 | Speech communication is most common in human beings. Currently, Taiwanese or Min-Nan spoken language is used by more than 75% of the people in Taiwan and approximated 50 million populations in the world. Although several written forms of Taiwanese have been proposed and used, there is lack of modem DSP research and development in Taiwanese speech communication. This has led to the detriment of Taiwanese culture, education and training of native language, and communication-aided technology development for the disabled. This study is aimed to develop a Taiwanese syllable-based synthesis units database by applying phonology' principles and DSP. Instead of intonation pronunciation, Taiwanese is a tone language with basic vowels, consonants, fundamental tone and other 7 tones. The single-tone database, consisting of 522 syllables, is determined from syllables screening by using the existed Chinese characters mapping to Taiwanese spoken words. Then an infinite-vocabulary TTS synthesis system is developed by applying DSP transformation of single-tone to other 7 tones on the database. An approach for the storage of corresponding acoustic waveform database has been through recording the synthesis units with syllable-based embedded sentences (SBES). Selection criteria ofSBES are based on a set of syllables consisting of the desired syllables without regarding to the semantics of sentences, and clear speech signal boundary for syllables' segmentation. During the recording process, speed, frequency, duration, and intensity of synthesis units are made as consistent as possible. The results of 522 single-tone syllables database have been shown to include synthesis units for all of the possible Taiwanese semantic syllables. The established infinite-vocabulary Taiwanese TTS system would be valuable to both provision of fundamental research and scientific applications in Taiwanese language. This research should provide fundamental database for speech analysis, synthesis, and recognition in modem Taiwanese Language. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。