Université Rennes 2, UFR ALC, Rennes, France.
Laboratoire de Phonétique et Phonologie (LPP) UMR 7018, CNRS & Université Sorbonne Nouvelle, Institut de Linguistique et de Phonétique Générales et Appliquées, Paris, France.
Clin Linguist Phon. 2023 Sep 2;37(9):866-882. doi: 10.1080/02699206.2022.2092425. Epub 2022 Jul 12.
This study aimed to assess the intelligibility of so-called 'pseudo-whispered speech' (pseudo-WS), as produced without voice nor pulmonic airstream by some alaryngeal patients prior to rehabilitation. Several perception tests were submitted to three experienced clinicians and three naive listeners, relying on the speech samples of 20 French native speakers: 10 alaryngeal speakers, solely using pseudo-WS when starting speech therapy up to six months after total laryngectomy, and 10 control speakers, recorded in the closest speech mode available, whispered speech (WS). Experts were asked to identify consonants (C) in the /a/+C+/a/ context and to rate intelligibility, unintended additive noise, and fluency on a likert-scale, while naive listeners completed a quantitative test of intelligibility. Intelligibility of WS was found to be high, with scores ranging from 46.33/54 to 53.67/54 (median 52.5, interquartile range 2.33) for the quantitative test, and segmental intelligibility ranging from 68.75% to 94.79% (median 87.5, interquartile range 17.71). Segmental confusion affected voicing in favour of unvoiced consonants, as previously reported in the literature. By contrast, intelligibility of pseudo-WS was found to be poor, with scores ranging from 1/54 (unintelligibility) to 28.33/54 (median 8.66, interquartile range 14.67) for the quantitative test, and segmental intelligibility ranging from 3.13% to 28.13% (median 9.24, interquartile range 14.58). Segmental intelligibility was not uniformly affected: stops, labials and unvoiced consonants were better identified than other categories. Finally, a significant correlation was found between global intelligibility and articulatory precision, while unintended additive noise and fluency seemed to play no role.
本研究旨在评估某些喉切除患者在康复前使用的所谓“伪耳语”（pseudo-WS）的可懂度。该研究向三位经验丰富的临床医生和三位非专业听众提供了几种感知测试，测试对象是 20 位法国母语者的语音样本：10 位喉切除患者，在接受喉切除术后的 6 个月内仅使用伪耳语进行言语治疗；10 位对照者，采用最接近的言语模式录制，即耳语（WS）。专家被要求识别 /a/+C+/a/ 语境中的辅音（C），并根据可懂度、意外附加噪声和流利程度在李克特量表上进行评分，而非专业听众则完成了一个可懂度的定量测试。WS 的可懂度被发现很高，定量测试的分数范围为 46.33/54 到 53.67/54（中位数为 52.5，四分位距为 2.33），分段可懂度的分数范围为 68.75%到 94.79%（中位数为 87.5，四分位距为 17.71）。分段混淆影响了语音，有利于无语音辅音，这与文献中的先前报道一致。相比之下，pseudo-WS 的可懂度被发现很差，定量测试的分数范围为 1/54（不可懂）到 28.33/54（中位数为 8.66，四分位距为 14.67），分段可懂度的分数范围为 3.13%到 28.13%（中位数为 9.24，四分位距为 14.58）。分段可懂度并不是均匀受影响的：塞音、唇音和无语音辅音比其他类别更容易被识别。最后，发现整体可懂度与发音准确度之间存在显著相关性，而意外附加噪声和流利程度似乎没有作用。
