背景:言语和语言治疗师(SLT)定期使用语音转录来记录和分析典型和无序的言语。语音转录对听觉感知技能的要求很高,因此研究人员对其准确性和可靠性持怀疑态度。文献描述了语音转录如何容易产生听觉错觉和偏见,例如,偏好从转录者自己的语言中转录语音。很少有实证研究计算转录者之间的协议,其中已经报告了一系列协议得分(51%-97%)。人们一致认为,随着语音细节的增加,同意率降低。元音和辅音在国际音标(IPA)中具有不同的感知特征,因此它们的协议率可能不同,到目前为止,有矛盾的证据,是否元音或辅音更多的同意。迄今为止,转录协议研究最常招募语音专家而不是SLT,因此需要进一步研究以确定SLT之间的转录协议及其对临床实践的影响。
目的:该研究的主要目的是计算一组讲英语的SLT的协议分数,这些SLT在生态有效的环境中转录了无序的语音样本。该研究还试图通过比较元音或辅音是否可以达成更多共识,从而为先前存在的矛盾证据基础做出贡献。研究人员旨在根据使用频率和使用一致性来评论SLT对变音符号和非母语语音符号(英语语音清单中未包含的符号)的使用。通过分析转录,该研究旨在讨论转录变异性对语音错误模式的影响,从而对诊断等临床决策的影响,干预和治疗目标的选择。
方法:通过两个国家健康服务信托机构的便利样本招募了12名儿科SLT,其中两名是语音障碍(SSD)专家。参与者从一个说话混乱的男孩的视频中转录了16个单词,完成了对发音和语音的诊断评估(DEAP,多德等人。,2006)来自远程医疗任命。手动分析了变音符号和非英语IPA符号的使用。使用Python脚本为每个目标单词计算了经典的一致性分数,然后是孤立的元音和辅音。手动分析数据集,以确定转录差异是否导致识别不同的语音错误模式。研究人员认为这在临床实践中的意义。
结果:平均经典协议分数为56.3%。辅音比元音更同意,同意分数为62.8%和48.6%,分别。9名参与者(75%)使用变音符号(最常见的是长度标记),8名参与者(67%)在其数据集中至少一次转录了非本地IPA符号。但通常频率低,使用之间也有协议。转录捕获了典型和非典型错误模式的发生,但只有三种错误模式,在确定的20个中,出现在所有12名参与者的转录中。
结论:56.3%的一致性分数质疑SLT中转录的准确性和可靠性,这是该行业的一项基本技能。研究结果表明,考虑到较低的同意率,SLT在解释元音时要比辅音更加谨慎。非本地符号和变音符号的使用频率相对较低,这可能反映出它们的使用精度较低或转录这些符号的置信度降低。该研究讨论了转录变化如何影响语音和语音分析,这反过来会影响临床决策,如诊断SSD,选择进一步的诊断评估,选择治疗目标和干预措施。皇家言语和语言治疗师学院认可的转录指南(儿童言语障碍研究网络,2017)可以进行修订,以传达对SLT转录技能的现实期望,或SLT应提供更多的培训,以提高转录技能,以满足当前的期望。讨论了其他提高转录准确性的建议,例如通过仪器方法,然而,这些都有自己的局限性,如实用性,费用和专业培训的需求。
结论:在这个主题上已经知道语音转录对人类感知技能的要求很高,研究人员对其可靠性表示怀疑。很少有实证研究计算转录者之间的一致性,根据研究条件,已经报告了一系列协议分数(51%-97%)。研究主要涉及经验丰富的转录者(例如,语音治疗师),而不是言语和语言治疗师(SLT),一个期望定期使用语音转录来记录和分析典型和无序语音的职业。这项研究对现有知识的补充在以前的研究中已经报道了一系列转录协议分数,主要比较成对或小组的专家转录者,而不是SLT。当一组12个SLT在生态有效的环境中转录无序的语音样本时,这项研究提供了56.3%的一致性分数(其中语音样本是从使用诊断评估的远程健康预约中进行的真实语音评估中获取的语音样本)。研究发现辅音比元音更受欢迎,增加了矛盾的证据基础。与其他研究不同,研究人员分析转录以确定错误模式,以检查转录变异对临床决策的影响。这项工作的潜在或实际临床意义是什么?研究人员质疑SLT是否符合皇家言语和语言治疗师学院(RCSLT)认可的转录指南(儿童言语障碍研究网络,2017年),协议得分相对较低,为56.3%。该研究还质疑使用变音符号和非英语国际音标(IPA)符号的可靠性,以及由于感知限制,是否应期望SLT使用这些符号。数据集中出现了20个语音过程,在所有12名参与者中,只有3人达成了协议。因此,研究人员讨论了转录变异如何导致不同的诊断,治疗目标和干预措施的选择。该论文建议需要更多的培训来提高转录的准确性,并考虑了使用仪器方法的适当性,同时认识到其局限性,如可行性,成本和专业培训需求。
BACKGROUND: Speech and language therapists (SLTs) regularly use phonetic transcription to record and analyse typical and disordered speech. Phonetic transcription is highly demanding of auditory perceptual skills so researchers are sceptical about its accuracy and reliability. The literature describes how phonetic transcription is prone to auditory illusions and biases, such as a preference to transcribe speech sounds from the transcriber\'s own language. Few empirical research studies have calculated agreement amongst transcribers where a range of agreement scores have been reported (51%-97%). There is a consensus that agreement rates decrease as phonetic detail increases. Vowels and consonants are characterised by different perceptual features within the International Phonetic Alphabet (IPA) so they may differ in agreement rates, and thus far there is contradictory evidence as to whether vowels or consonants are more agreed upon. Transcription agreement studies to date have most commonly recruited phoneticians rather than SLTs so further research is warranted to determine transcription agreement amongst SLTs and its impact on clinical practice.
OBJECTIVE: The study\'s primary aim was to calculate agreement scores from a group of English-speaking SLTs who transcribed disordered speech samples in an ecologically valid setting. The study also sought to contribute to the pre-existing contradictory evidence base regarding whether vowels or consonants may be more agreed upon by comparing their agreement scores. The researcher aimed to comment on SLTs\' use of diacritics and non-native speech symbols (symbols not included in the English phonetic inventory) in terms of their frequency and agreement of use. By analysing transcriptions, the study aimed to discuss the impact transcription variability has on speech sound error patterns and thus its impact on clinical decision-making such as diagnosis, choice of intervention and therapy targets.
METHODS: Twelve paediatric SLTs were recruited via a convenience sample at two National Health Service trusts, two of whom were specialists in Speech Sound Disorders (SSDs). Participants transcribed 16 words from a video of a boy with disordered speech completing the Diagnostic Evaluation of Articulation and Phonology (DEAP, Dodd et al., 2006) from a telehealth appointment. The use of diacritics and non-English IPA symbols were manually analysed. A classic agreement score was calculated for each target word using a Python script, and then for vowels and consonants in isolation. Datasets were manually analysed to determine whether differences in transcription resulted in the identification of different speech sound error patterns. The researcher considered the implications this had within clinical practice.
RESULTS: The average classic agreement score was 56.3%. Consonants were more agreed upon than vowels with agreement scores of 62.8% and 48.6%, respectively. Nine participants (75%) used diacritics (most commonly length marks) and eight participants (67%) transcribed non-native IPA symbols at least once in their datasets, but generally with low frequencies and agreements amongst their use. Transcriptions captured the occurrence of typical and atypical error patterns but only three error patterns, out of the 20 identified, were present in all 12 participants\' transcriptions.
CONCLUSIONS: The agreement score of 56.3% questions the accuracy and reliability of transcription amongst SLTs which is an essential skill of the profession. The findings highlight SLTs should be more cautious of interpreting vowels than consonants given lower agreement rates. The frequency of use of non-native symbols and diacritics was relatively low which could reflect a low accuracy of their use or reduced confidence in transcribing these. The study discussed how variations in transcriptions can impact phonological and phonetic analysis, which in turn can influence clinical decision-making such as diagnosing SSDs, selecting further diagnostic assessments and choosing therapy targets and interventions. The Royal College of Speech and Language Therapists-endorsed transcription guidelines (Child Speech Disorder Research Network, 2017) could be revised to convey realistic expectations of SLTs\' transcription skills, or SLTs should be offered more training to improve transcription skills to meet current expectations. Other suggestions to improve transcription accuracy are discussed such as via instrumental methods, yet these come with their own limitations such as practicality, costs and need for specialist training.
CONCLUSIONS: What is already known on this subject Phonetic transcription is highly demanding of human perceptual skills, and researchers are sceptical about its reliability. There are few empirical research studies calculating agreement amongst transcribers, and a range of agreement scores have been reported (51%-97%) dependent on the research conditions. Research mostly involves experienced transcribers (e.g., phoneticians) rather than speech and language therapists (SLTs), a profession expected to regularly use phonetic transcription to record and analyse typical and disordered speech. What this study adds to existing knowledge A range of transcription agreement scores have been reported in previous studies, mainly comparing pairs or small groups of specialist transcribers rather than SLTs. This study provides an agreement score of 56.3% when a group of 12 SLTs transcribed a disordered speech sample in an ecologically valid setting (where speech samples were taken from a real-life speech sound assessment over a telehealth appointment using the Diagnostic Evaluation of Articulation and Phonology). The study found consonants are more agreed upon than vowels, adding to the contradictory evidence base. Unlike other studies, the researcher analysed transcriptions to identify error patterns to examine the impact that transcription variation has on clinical decision-making. What are the potential or actual clinical implications of this work? The researcher questions whether SLTs are meeting the expectations of \'accurate transcription\' as listed by the Royal College of Speech and Language Therapists (RCSLT) endorsed transcription guidelines (Child Speech Disorder Research Network, 2017) given the relatively low agreement score of 56.3%. The study also questions the reliability of the use of diacritics and non-English International Phonetic Alphabet (IPA) symbols and whether SLTs should be expected to use these due to perceptual limitations. Twenty phonological processes emerged from the datasets, only three of which were agreed upon across all 12 participants. The researcher therefore discusses how transcription variations could result in different diagnoses, therapy targets and interventions choices. The paper suggests more training is required to enhance transcription accuracy, and also considers the appropriateness of utilising instrumental methods whilst recognising its limitations such as feasibility, costs and specialist training needs.