華語中說謊的聲學及電聲門訊號特徵

Abstract

本研究探討華語中說謊的聲學及電聲門訊號特徵，並且有三大研究重點：（ㄧ）說謊時的聲學及電聲門整體特徵、（二）性別對於聲學及電聲門特徵的影響以及（三）成功/失敗的謊話及實話的特徵。研究參與者為三十六位華語母語者（十八位女性）。每位研究參與者皆參加由十四個生平問題組成的互動式訪談問答，並且由另外兩位實驗人員來進行訪談。研究參與者必須選擇在一半的問題中說謊，而在另一半的問題中說真話。研究參與者在回答問題時，聲學及電聲門的訊號同時進行錄製。研究參與者的回答條件以及實驗人員的知覺條件皆有紀錄起來方便後續分析使用。本研究共使用了二十種聲學參數［包含基頻（F0）相關參數、聲音強度（intensity）相關參數、第一至第五共振峰（formant）、頻率擾動度（jitter）、音量擾動度（shimmer）、諧噪比（HNR）、第一及第二諧波差（H1-H2）、語速］以及十四種電聲門參數［包含導數法接觸商（CQDEGG）、混合法接觸商（CQMIX）、接觸時間商（contacting-time quotient）、分離時間商（decontacting-time quotient）、中坡法速度商（speed quotient with a mid-slope criterion）、接觸指數（contact index）、基頻相關參數、頻率擾動度、音量擾動度、諧噪比］。研究結果顯示中文說謊的整體特徵包含基頻平均值、聲音強度平均值、第一共振峰、第五共振峰、接觸商、分離時間商、接觸指數等數值上升，以及頻率擾動度、音量擾動度、諧噪比、第四共振峰等數值下降。此結果與過往的文獻有相當程度的呼應，並意味著中文的謊言可能與特定的語用功能（如：聲音聽起來更堅定及好聽）、較強的情緒激發、聲門接觸對稱性上升、較為受壓的發聲等特質有關。此外，中文說謊同時存在跨性別和特定性別特徵。跨性別特徵包含聲音強度平均值、第五共振峰上升以及頻率擾動度、諧噪比、第四共振峰下降，更加突顯兩性皆會使用特定語用功能（如：試圖聽起來更堅定及好聽）的可能性。特定性別特徵則包含女性說謊時會在基頻平均值、音量擾動度、第一共振峰、第三共振峰、分離時間商等數值上升以及在基頻最大值、接觸商等數值下降。男性則有接觸商、接觸指數上升以及音量擾動度下降等特徵。此結果顯示女性在說謊時可能較容易有情緒激發及構音上的改變，而兩性在發聲特質上可能有相反的變化趨勢（亦即女性有較輕的發聲、男性有較受壓的發聲）。最後，失敗謊話與成功謊話相較之下有基頻平均值、頻率擾動度、音量擾動度上升以及基頻最大值、分離時間商下降的特徵，代表其有音高較高、聲音較不好聽、語調峰值較低、聲帶分離時間較短等特質。失敗真話與成功真話相較之下則有分離時間商、第一共振峰上升以及第三共振峰下降的特徵，顯示其構音相關之頻譜特質及較長的聲帶分離時間。此結果揭露華語使用者將謊話及真話視為真或假的可能判斷依據。This study examined the acoustic and electroglottographic (EGG) profile of Mandarin deception with three major focuses: (1) the global acoustic and EGG characteristics of Mandarin deception, (2) the influence of gender and (3) characteristics of successful/ unsuccessful lies and truths. Thirty-six Mandarin speakers (18 females) participated in an interactive interview game, where they were asked 14 biographical questions by two additional experimenters. Participants were instructed to choose half of the questions for deceptive answers and the other half for truthful answers. Acoustic and EGG signals of the participants’ responses were simultaneously recorded. Conditions (deception& truth) of both participants’ production and experimenters’ perception were noted for analyses. Analyses included 20 acoustic parameters [i.e., F0-related measures, intensity-related measures, formants (F1-F5), jitter, shimmer, harmonics-to-noise ratio (HNR), H1-H2, and speech rate] and 14 EGG parameters [i.e., CQDEGG, CQMIX, contacting-time quotient (CTQ), decontacting-time quotient (DTQ), speed quotient with a mid-slope criterion (SQ-mid), contact index (CI), F0-related measures, jitter, shimmer, and HNR]. Increases in F0 mean, intensity mean, F1, F5, CQ, DTQ, and CI as well as decreases in jitter, shimmer, HNR, F4 significantly correlated with deception across gender. These features suggest Mandarin deception could be tied to specific pragmatic functions (e.g., sounding firmer and more pleasant), greater emotional arousal, increased glottal contact symmetry and more pressed phonation, which well echoed findings from the literature. Additionally, both cross-gender and gender-specific features exist in Mandarin deception. Cross-gender features included increases in intensity mean and F5 and decreases in jitter, HNR and F4, which further strengthened the idea that certain pragmatic functions (e.g., trying to sound firmer and more pleasant) are utilized by both genders. Gender-specific features encompassed increases in F0 mean, shimmer, F1, F3, and DTQ as well as decreases in F0 maximum and CQ for female deception and increases in CQ and CI and decreases in shimmer for male deception, showing females are more prone to changes in emotional arousal and articulation and that both genders could undergo opposing changes in phonation quality (i.e., lighter registers for females and heavier registers for males). Furthermore, unsuccessful lies differed from their successful counterparts by increases in F0 mean, jitter and shimmer as well as decreases in F0 maximum and DTQ, showing qualities like higher F0 levels, less pleasant voice qualities, compromised intonation peaks, and shorter vocal fold decontacting periods. Unsuccessful truths, on the other hand, manifested increases in DTQ and F1 and decreases in F3, denoting spectral properties pertaining to articulation and longer vocal fold decontacting periods. The results could bring to light the mechanisms behind how Mandarin speakers perceive lies and truths as deceptive or truthful.