タイ語の規則合成に用いるデミシラブルの孤立発声単音節単語からの自動抽出の一方法
スポンサーリンク
概要
- 論文の詳細を見る
In this paper, we describe a procedure for automatically extracting demisyllables(DSs) from Thai isolated source words of the forms consonant-vowel, vowel-consonant, and vowel. Routines that are particular to each category of source words are used to establish DS boundary locations. All segmental cues are derivable from the cepstrum represennting a log spectral envelope that approximately passes through peaks of the fine structure, which have been manually determined based on spoken source words of one male talker. The automatic procedure has been used to extract DSs from spoken source words of two new talkers:one male and one female. Manual comparisons have shown that automatically obtained DS boundary locations were within three frames(1 frame is 10 ms) from the corresponding locations obtained manually by a trained native linguist. For the male talker, 94. 2% of the automatically obtained locations coincided with the corresponding manually obtained locations, and 3. 5%, 1. 9%, and 0. 4% differed by one, two, and three frames, respectively; for the female talker, 95. 7% coincided, and 3. 3%, 0. 8%, and 0. 2% differed by one, two, and three frames, respectively. The above results are comparable with the results of using the automatic procedure to extract the DSs from the source words of the old talker:97. 5% coincided, and 2. 3% and 0. 2% differed by one and two frames, respectively. These results indicate that the automatic procedure promises a very good first estimate set of DSs.
- 社団法人日本音響学会の論文
著者
-
今井 聖
Research Laboratory Of Precision Machinery And Electronics Tokyo Institute Of Technology
-
チャッチャバリ サラバリ
Research Laboratory of Precision Machinery and Electronics, Tokyo Institute of Technology
-
チャッチャバリ サラバリ
Research Laboratory Of Precision Machinery And Electronics Tokyo Institute Of Technology