- 作者: 曾進興
- 作者服務機構: 國立高雄師範大學特殊教育系
- 中文摘要: 在連續語音的辨識過程中,斷詞是一項極為棘手的問題,原因是語音當中缺乏一致而有效的聲學線索,可為聽話者用來確定詞界的位置。然而,聽話者確實有能力把語音切割成某種單位,否則詞彙檢索就難以進行。文獻上已有若干語音辨識模型提到切割程序。漢語使用者如何切音,曾進興等人在先前的文章中已有討論。曾進興及其合作者主張漢語使用者是在音節這個層次上進行切音的,理由是:音節比起音素或單詞都有著較鮮明的聲學標幟;在數量上音節比單詞少得太多了,因此在記憶的搜尋上音節的負荷輕了許多;另外,中文書寫系統中「一字一音」的原則,恐怕不能說毫無知覺心理上的根源。這個想扶並且也得到了幾個音素偵測實驗的支持,這些實驗的結果發現空缺劣勢效果、字調效果及音節性效果。本研究的目的是想:(1)使用「超語言判斷」探討重音音步是否在理論上為一可行的切音策略(2)使用兩種實驗策略探討字調效果的來源;以及(3)使用臺灣話來覆証字調效果和音節複雜度效果(不存在)。實驗結果發現:(1)國語重音分派原則並不明顯故而否定重音切割的可能性;(2)字調效果可能起源於音節長度的變異;以及(3)臺灣話也可觀察到字調效果,但音節複雜度同樣不起任何作用。這些發現可以視為對音節切割假說的進一步支持。
- 英文摘要: Word segmentation is one of the most difficult tasks in continuous speech recognition. The problemarises because there do not appear to be consistent and distinct acoustic cues that can be employed by thelisteners for locating word boundaries. Even so, listeners are apparently able to parse in one way or anotherthe time-varying acoustic signal into word-like units before the lexical matching process is completed. Variousexplicit and implicit segmentation procedures have been proposed in the literature. How the Chinese spokenlanguage is segmented by native speakers has been addressed in two publications by the author and coworkers.We have argued that syllable is the level at which Chinese speakers segment their spoken language. Theargument was made on the guounds that syllables have more reliable acoustic fingerprint than phonemes orwords, that they are much less in number than words and that the Chinese orthography enjoys a one-character-one-syllable principle. This theoretical position was substantiated by findings from a series of phoneme-identification experiments. Major findings from previous studies included the gap inferiority effect, the toneeffect, and the syllabicity effect. This study was designed (1) to investigate the plausibility of stress foot beinga segmentation unit for Chinese speakers, (2) to discover the locus of the tone effect and (3) to replicate thetone effect and the (absence of)syllable complexity effect with Taiwanese speech materials. The resultsindicated that the stress assignment in Mandarin is not consistent enough to warrant a stress segmentationprocedure, that the tone effect arises as a result of syllabic length variations, and that the tone effect was, butthe syllable complexity was not, observed in Taiwanese. These findings are considered to support the syllabicsegmentation hypothesis proposed in earlier studies.
- 中文關鍵字: 音節; 語音切割; 語音辨識
- 英文關鍵字: syllable; speech segmentation; speech recognition