Transcription Accuracy Research Articles

ClinClip: a Multimodal Language Pre-training model integrating EEG data for enhanced English medical listening assessment.

In the field of medical listening assessments,accurate transcription and effective cognitive load management are critical for enhancing healthcare delivery. Traditional speech recognition systems, while successful in general applications often struggle in medical contexts where the cognitive state of the listener plays a significant role. These conventional methods typically rely on audio-only inputs and lack the ability to account for the listener's cognitive load, leading to reduced accuracy and effectiveness in complex medical environments. To address these limitations, this study introduces ClinClip, a novel multimodal model that integrates EEG signals with audio data through a transformer-based architecture. ClinClip is designed to dynamically adjust to the cognitive state of the listener, thereby improving transcription accuracy and robustness in medical settings. The model leverages cognitive-enhanced strategies, including EEG-based modulation and hierarchical fusion of multimodal data, to overcome the challenges faced by traditional methods. Experiments conducted on four datasets-EEGEyeNet, DEAP, PhyAAt, and eSports Sensors-demonstrate that ClinClip significantly outperforms six state-of-the-art models in both Word Error Rate (WER) and Cognitive Modulation Efficiency (CME). These results underscore the model's effectiveness in handling complex medical audio scenarios and highlight its potential to improve the accuracy of medical listening assessments. By addressing the cognitive aspects of the listening process. ClinClip contributes to more reliable and effective healthcare delivery, offering a substantial advancement over traditional speech recognition approaches.

A Researcher’s Journey from Traditional Qualitative Methods to Tech-Driven Insights

The integration of technology into qualitative research has revolutionised traditional methods, providing researchers with new opportunities and challenges. This paper compares the effectiveness, strengths, and limitations of traditional versus technology-based methods for recruiting participants, conducting interviews, and transcribing data, based on the personal experiences of the researchers. The findings reveal that while traditional face-to-face interviews enable deeper personal interaction and richer data collection, they are often time-consuming and logistically challenging. In contrast, technology-based methods offer increased accessibility, efficiency, and flexibility, particularly when recruiting participants and conducting interviews among geographically dispersed populations. However, these methods may compromise the depth of data due to the challenges of maintaining non-verbal cues and transcription accuracy, particularly when using automated transcription tools. This paper recommends a hybrid approach that combines the strengths of both methods. Best practices for using technology in qualitative research are proposed to ensure the richness and contextual depth of the data while leveraging the benefits of modern tools.

Older adults’ recognition of medical terminology in hospital noise

Word identification accuracy is modulated by many factors including linguistic characteristics of words (frequent vs. infrequent), listening environment (noisy vs. quiet), and listener-related differences (older vs. younger). Nearly, all studies investigating these factors use high-familiarity words and noise signals that are either energetic maskers (e.g., white noise) or informational maskers composed of competing talkers (e.g., multitalker babble). Here, we expand on these findings by examining younger and older listeners’ speech-in-noise perception for words varying in both frequency and familiarity within a simulated hospital noise that has important non-speech information. The method was inspired by the real-world challenges aging patients can face in understanding less familiar medical terminology used by healthcare professionals in noisy hospital environments. Word familiarity data from older and young adults were collected for 800 medically related terms. Familiarity ratings were highly correlated between the two age groups. Older adults’ transcription accuracy for sentences with medical terminology that vary in their familiarity and frequency was assessed across four listening conditions: hospital noise, speech-shaped noise, amplitude-modulated speech-shaped noise, and quiet. Listeners were less accurate in noise conditions than in a quiet condition and were more impacted by hospital noise than either speech-shaped noise. Sentences with low-familiarity and low-frequency medical words combined with hospital noise were particularly detrimental for older adults compared to younger adults. The results impact our theoretical understanding of speech perception in noise and highlight real-world consequences of older adults’ difficulties with speech-in-noise and specifically noise containing competing, non-speech information.

Decoding disparities: evaluating automatic speech recognition system performance in transcribing Black and White patient verbal communication with nurses in home healthcare.

As artificial intelligence evolves, integrating speech processing into home healthcare (HHC) workflows is increasingly feasible. Audio-recorded communications enhance risk identification models, with automatic speech recognition (ASR) systems as a key component. This study evaluates the transcription accuracy and equity of 4 ASR systems-Amazon Web Services (AWS) General, AWS Medical, Whisper, and Wave2Vec-in transcribing patient-nurse communication in US HHC, focusing on their ability in accurate transcription of speech from Black and White English-speaking patients. We analyzed audio recordings of patient-nurse encounters from 35 patients (16 Black and 19 White) in a New York City-based HHC service. Overall, 860 utterances were available for study, including 475 drawn from Black patients and 385 from White patients. Automatic speech recognition performance was measured using word error rate (WER), benchmarked against a manual gold standard. Disparities were assessed by comparing ASR performance across racial groups using the linguistic inquiry and word count (LIWC) tool, focusing on 10 linguistic dimensions, as well as specific speech elements including repetition, filler words, and proper nouns (medical and nonmedical terms). The average age of participants was 67.8 years (SD = 14.4). Communication lasted an average of 15 minutes (range: 11-21 minutes) with a median of 1186 words per patient. Of 860 total utterances, 475 were from Black patients and 385 from White patients. Amazon Web Services General had the highest accuracy, with a median WER of 39%. However, all systems showed reduced accuracy for Black patients, with significant discrepancies in LIWC dimensions such as "Affect," "Social," and "Drives." Amazon Web Services Medical performed best for medical terms, though all systems have difficulties with filler words, repetition, and nonmedical terms, with AWS General showing the lowest error rates at 65%, 64%, and 53%, respectively. While AWS systems demonstrated superior accuracy, significant disparities by race highlight the need for more diverse training datasets and improved dialect sensitivity. Addressing these disparities is critical for ensuring equitable ASR performance in HHC settings and enhancing risk prediction models through audio-recorded communication.

Аналіз сучасних підходів до розв’язання задач розпізнавання мовлення

Introduction. The necessity for modern approaches to solving speech recognition tasks arises from the rapid development of artificial intelligence and the need to improve the accuracy and speed of human-computer interaction in various areas, such as voice assistants, translation, and automation. This direction is becoming increasingly relevant due to the growing volume of generated audio data and the need for real-time processing, particularly in Ukrainian contexts where multiple languages and dialects coexist. Currently, several approaches to speech recognition, analysis, and transcription exist, including methods based on neural networks, speaker diarization techniques, noise removal, and data structuring. However, the challenge of creating a universal solution that meets the needs of multilingual environments and effectively handles unstructured audio data remains relevant. Purpose. To review existing tools and algorithms for solving speech recognition tasks, particularly for Ukranian. Methods. Speech recognition, deep learning, transformers. Results. Theoretical foundations of approaches and models for speech recognition were considered for building a knowledge base for a multilingual spoken dialogue system. Effective examples of improving transcription accuracy for languages with limited data were also explored, along with potential steps to enhance system speed. Potential datasets for model training were discussed. Conclusion. A structured review of modern methods for processing and analyzing multilingual audio files was provided, outlining their advantages, disadvantages, and unresolved issues.

Advancing Healthcare: Intelligent Speech Technology for Transcription, Disease Diagnosis, and Interactive Control of Medical Equipment in Smart Hospitals

Intelligent Speech Technology (IST) is revolutionizing healthcare by enhancing transcription accuracy, disease diagnosis, and medical equipment control in smart hospital environments. This study introduces an innovative approach employing federated learning with Multi-Layer Perceptron (MLP) and Gated Recurrent Unit (GRU) neural networks to improve IST performance. Leveraging the “Medical Speech, Transcription, and Intent” dataset from Kaggle, comprising a variety of speech recordings and corresponding medical symptom labels, noise reduction was applied using a Wiener filter to improve audio quality. Feature extraction through MLP and sequence classification with GRU highlighted the model’s robustness and capacity for detailed medical understanding. The federated learning framework enabled collaborative model training across multiple hospital sites, preserving patient privacy by avoiding raw data exchange. This distributed approach allowed the model to learn from diverse, real-world data while ensuring compliance with strict data protection standards. Through rigorous five-fold cross-validation, the proposed Fed MLP-GRU model demonstrated an accuracy of 98.6%, with consistently high sensitivity and specificity, highlighting its reliable generalization across multiple test conditions. In real-time applications, the model effectively performed medical transcription, provided symptom-based diagnostic insights, and facilitated hands-free control of healthcare equipment, reducing contamination risks and enhancing workflow efficiency. These findings indicate that IST, powered by federated neural networks, can significantly improve healthcare delivery, accuracy in patient diagnosis, and operational efficiency in clinical settings. This research underscores the transformative potential of federated learning and advanced neural networks for addressing pressing challenges in modern healthcare and setting the stage for future innovations in intelligent medical technology.

Automated speech recognition bias in personnel selection: The case of automatically scored job interviews.

Organizations, researchers, and software increasingly use automatic speech recognition (ASR) to transcribe speech to text. However, ASR can be less accurate for (i.e., biased against) certain demographic subgroups. This is concerning, given that the machine-learning (ML) models used to automatically score video interviews use ASR transcriptions of interviewee responses as inputs. To address these concerns, we investigate the extent of ASR bias and its effects in automatically scored interviews. Specifically, we compare the accuracy of ASR transcription for English as a second language (ESL) versus non-ESL interviewees, people of color (and Black interviewees separately) versus White interviewees, and male versus female interviewees. Then, we test whether ASR bias causes bias in ML model scores-both in terms of differential convergent correlations (i.e., subgroup differences in correlations between observed and ML scores) and differential means (i.e., shifts in subgroup differences from observed to ML scores). To do so, we apply one human and four ASR transcription methods to two samples of mock video interviews (Ns = 1,014 and 414), and then we train and test models using these different transcripts to score multiple constructs. We observed significant bias in the commercial ASR services across nearly all comparisons, with the magnitude of bias differing across the ASR services. However, the transcription bias did not translate into meaningful measurement bias for the ML interview scores-whether in terms of differential convergent correlations or means. We discuss what these results mean for the nature of bias, fairness, and validity of ML models for scoring verbal open-ended responses. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

From Rainbow to Rambo: Speaking to an AI and the Taiwanese English Pronunciation Intelligibility

The purpose of the study was to investigate the pronunciation intelligibility of Taiwan English (TE). An artificial-intelligence (AI)-enhanced automatic speech recognition (ASR) software program was used to perform a speech-to-text transcription, which enabled voice recognition algorithms to automatically produce orthographic transcription(s) of the Taiwanese university students’ speech samples. The findings based on the intelligibility scores and the comparison of the phonetic discrepancies of the transcribed words suggest that TE appears to be highly intelligible, however, it was clear that sub-varieties of TE exist, bordering from basilect (perceived to be less prestigious) to the cultivated/educated acrolect form, which is closer to the standard American English (SAE), thus making it even more challenging to identify a standardized TE. Nonetheless, the (in)accuracy of the ASR transcription gave the students the opportunity to reflect on their English pronunciation, to consider how comprehensible they sound, and to strive "to own" their English as part of their Taiwanese identity.

Artificial intelligence powered OCR models for digitizing ayurveda manuscripts

This article presents a novel approach to transcribe Ayurveda manuscripts using Optical Character Recognition (OCR) technology enhanced with machine learning and forensic handwriting analysis techniques. Traditional transcription methods for these manuscripts are time-consuming and challenging due to diverse scripts and handwriting styles. The proposed methodology involves collecting manuscript samples from various sources and training OCR models to recognize the nuances of these scripts. By incorporating forensic handwriting principles, such as consistency and natural variation, the OCR models achieve greater accuracy. Additionally, the development of user-friendly interfaces allows scholars to train these models without needing technical expertise, streamlining the digitization process. Preliminary results show that this approach significantly improves transcription accuracy compared to conventional OCR software. This method offers a scalable solution for digitizing Ayurveda texts, preserving cultural heritage, and accelerating research in Ayurveda medicine. Future research will focus on refining OCR models and expanding their application to other scripts and languages.

Learner training for phonetic transcription of typical and/or disordered speech: A scoping review.

Phonetic transcription is a core skill of speech and language therapists/pathologists (SLT/Ps) for clinical assessment of speech sounds and/or errors, and linguists for investigation of phonetic phenomena in various languages; hence, it is included in the curriculum of the corresponding degree programme. Many experts and course instructors have discussed their opinions on different aspects of phonetic transcription teaching and/or reported their own training programmes. However, no review has systematically summarized the types of expert recommendations and training methodology reported in the literature. Such information is important for mapping the knowledge gaps, refining current teaching and planning further research. To systematically summarize (1) the materials and procedures for delivering and evaluating phonetic transcription training programmes, and (2) the opinions from experts in phonetic transcription teaching regarding but not limited to the content, rationale(s), format and structure, and timing of teaching phonetic transcription of typical speech and/or disordered speech reported in the literature. A scoping review was carried out by following the Joanna Briggs Institute methodology. PubMed, EBSCOhost and Web of Science were searched and citation searching of included papers was completed. The included papers were divided into training programme reports, of which data (e.g., type and number of speech stimuli used, type and number of learners, outcome measures) were charted, and expert opinion papers, analysed using content analysis. A total of 565 studies were retrieved. After excluding duplicates and irrelevant papers and merging two sources that reported the same training, a total of 23 sources on training programmes and six opinion papers were analysed. Most training were for English and for SLT/P students, with some for the linguistics students. There are variations in the training content (with phonetic transcription practice as the main procedure), delivery mode (some presented via websites or computer software), stimuli (audio recordings of typical adults and children with speech disorders were most used), feedback (mostly immediate feedback using answer keys) and outcome measures (mainly transcription accuracy of learners and user opinions). Content analysis of opinion papers determined five main categories: rationale for training; aspects of the delivery of training; transcription; problem areas noted; and strategies/resources. Implications and considerations for teaching are discussed, along with recommendations for research in the design of evidence-based training. The findings can contribute to the development of general guidelines about phonetic transcription training in educational programmes and the establishment of pre-registration competencies. What is already known on this subject Phonetic transcription is a core skill of SLT/Ps and linguists for clinical speech assessment and research, and it is an important component in the curriculum of the relevant degree programmes. Many experts and course instructors have discussed their opinions regarding phonetic transcription teaching and/or reported their own training programmes in various sources such as books and book chapters, research papers and conference proceedings. What this paper adds to the existing knowledge The paper reports a scoping review that systematically summarized the materials and procedures for delivering and evaluating phonetic transcription programmes and experts' opinions and recommendations for phonetic transcription teaching, reported in the literature. The data from these two types of sources allowed an integrated discussion of the various aspects of phonetic transcription training, such as content, format and structure, materials and timing of teaching and the evaluation of training efficacy. What are the potential or actual clinical implications of this work? The data synthesized and discussed provide a foundation for a review of current practice in phonetic transcription teaching in SLT education. The findings can contribute to the development of general guidelines about phonetic transcription training in educational programmes and the establishment of pre-registration competencies.

Exploring the Experiences, Perceptions and Social Dynamics of Electronic Cigarette Users: A Qualitative Study.

Electronic cigarettes, or e-cigarettes, are rapidly gaining popularity throughout the globe as safer alternatives to traditional cigarette smoking. There are significant public health concerns due to the uncertainty of long-term health consequences. This study aims to examine the experiences, attitudes and social dynamics of e-cigarette users to provide guidance for public health policies and interventions. In this qualitative descriptive study, semi-structured interviews were conducted. Fifteen e-cigarette users were recruited using a snowball sampling strategy, where initial participants referred other suitable users. All participants had at least 6 months of regular e-cigarette use. Data were collected through face-to-face interviews with open-ended questions. The interviews were transcribed verbatim and reviewed by the researchers. This review identified four main themes and eight sub-themes. The four main themes identified as a result of the analysis are as follows: (1) initiation and usage experiences; (2) perceptions of health effects; (3) social and environmental factors and (4) addiction and future plans. Participants primarily began using e-cigarettes out of curiosity, due to the appealing fragrance and the belief that they were less harmful than traditional cigarettes. Usage patterns varied among participants, with some reporting minimal cravings in the early morning hours. Opinions on the long-term health effects of e-cigarettes were diverse, with many participants acknowledging uncertainty. Peers and family members perceived e-cigarettes as more socially acceptable than traditional smoking, resulting in fewer negative reactions. Finally, participants' levels of addiction and intentions to quit varied, with some expressing a desire to reduce or cease usage due to health concerns. The findings of this study indicate that the adoption of e-cigarettes is influenced by a confluence of factors, including curiosity, the perception of reduced damage in comparison to conventional cigarettes and social acceptance. Notwithstanding these claimed advantages, users have a diverse array of beliefs and understandings concerning the enduring health consequences of e-cigarettes. The presence of varying levels of addiction and the corresponding aspirations to cease the behaviour highlight the necessity for focused public health interventions. E-cigarette users were actively involved in this study, providing essential insights and feedback throughout. Their first-hand experiences shaped the interview guide and contributed to the identification of key themes. Participants also reviewed and confirmed the accuracy of the interview transcripts, ensuring the reliability of the data.

Biological effect of U(VI) exposure on lung epithelial BEAS-2B cells

In this study, the biological effects of U(VI) exposure on lung epithelial cells were investigated by MTT assay, immunofluorescence, flow cytometry, and Western blotting. U(VI)-induced stress triggers oxidative stress in cells, activates MAPK signaling pathways, and promotes inflammation. Additionally, U(VI) causes damage to the cell membrane structure and severe DNA injury, impacting the accuracy of transcription and translation. The results demonstrate that U(VI) exposure significantly inhibits cell proliferation and migration. This is attributed to the disruption of the PI3K/AKT/GSK-3β/β-catenin signaling pathway and the reduction in CyclinD1 expression, leading to a delayed cell cycle, decreased growth rate, mitochondrial damage, and reduced energy metabolism. This study provides a comprehensive understanding of the molecular mechanisms underlying uranium-induced cellular toxicity in lung epithelial cells.

Automated remote speech-based testing of individuals with cognitive decline: Bayesian agreement of transcription accuracy.

We investigated the agreement between automated and gold-standard manual transcriptions of telephone chatbot-based semantic verbal fluency testing. We examined 78 cases from the Screening over Speech in Unselected Populations for Clinical Trials in AD (PROSPECT-AD) study, including cognitively normal individuals and individuals with subjective cognitive decline, mild cognitive impairment, and dementia. We used Bayesian Bland-Altman analysis of word count and the qualitative features of semantic cluster size, cluster switches, and word frequencies. We found high levels of agreement for word count, with a 93% probability of a newly observed difference being below the minimally important difference. The qualitative features had fair levels of agreement. Word count reached high levels of discrimination between cognitively impaired and unimpaired individuals, regardless of transcription mode. Our results support the use of automated speech recognition particularly for the assessment of quantitative speech features, even when using data from telephone calls with cognitively impaired individuals in their homes. High levels of agreement were found between automated and gold-standard manual transcriptions of telephone chatbot-based semantic verbal fluency testing, particularly for word count.The qualitative features had fair levels of agreement.Word count reached high levels of discrimination between cognitively impaired and unimpaired individuals, regardless of transcription mode.Automated speech recognition for the assessment of quantitative and qualitative speech features, even when using data from telephone calls with cognitively impaired individuals in their homes, seems feasible and reliable.

Identifying Pronunciation Challenges for Cantonese EFL Learners: A Focus on Intelligibility

This study investigates the intelligibility of Cantonese-accented English by examining segmental pronunciation errors and their impact on native English listeners’ ability to accurately transcribe spoken sentences. Building on a previous production experiment, this research focuses on identifying which vowel and consonant mispronunciations most hinder intelligibility. Ten native English listeners transcribed 1,000 sentences produced by Cantonese speakers, and transcription accuracy was calculated for each target sound. The results reveal that dark /l/ (e.g., “cold”), final consonants like /g/ (e.g., “bag”), and the vowel /æ/ (e.g., “pan”) had the lowest accuracy rates, while sounds like /n/ (e.g., “no”) and /ʃ/ (e.g., “sheep”) were consistently transcribed correctly. These findings highlight the importance of targeting specific pronunciation difficulties in Cantonese EFL learners to improve intelligibility. The study supports the prioritization of intelligibility over accent reduction in pronunciation instruction, providing practical implications for teaching pronunciation in Cantonese-speaking contexts.

Optimising speech recognition using LLMs: an application in the surgical domain

Abstract Automatic speech recognition (ASR), powered by deep learning techniques, is crucial for enhancing humancomputer interaction. However, its full potential remains unrealized in diverse real-world environments, with challenges such as dialects, accents, and domain-specific jargon, particularly in fields like surgery, persisting. Here, we investigate the potential of large language models (LLMs) as error correction modules for ASR.We leverage Whisper-medium or ASRLibriSpeech for speech recognition, and GPT-3.5 or GPT-4 for error correction.We employ various prompting methods, from zero-shot to few-shot with leading questions and sample medical terms to correct wrong transcriptions. Results, measured by word error rate (WER), reveal Whisper’s superior transcription accuracy over ASR-LibriSpeech, with a WER of 11.93% compared to 32.09%. GPT-3.5, with the few-shot with medical terms prompting method, further enhances performance, achieving a 64.29% and 37.83% WER-reduction for Whisper and ASR-LibriSpeech, respectively. Additionally, Whisper exhibits faster execution speed. Substituting GPT-3.5 with GPT- 4 further enhances transcription accuracy. Despite some few challenges, our approach demonstrates the potential of leveraging domain-specific knowledge through LLM prompting for accurate transcription, particularly in sophisticated domains like surgery.

Supporting Inclusivity Through an Automatic Transcription Application to Improve Hearing Skills for the Deaf

Inclusivity is the principle that underlies a just and equal society for all individuals, including students or pupils who have hearing disabilities such as the deaf. In an effort to realize inclusivity, technological developments have opened the door to solutions that can improve the quality of life for deaf users. One of the main challenges faced by deaf users is difficulty in understanding and communicating with the outside world through hearing. This article discusses how automatic transcription applications can serve as powerful tools in supporting inclusiveness by improving the listening skills of deaf users. An automatic transcription application is software that is capable of converting spoken conversations into text in real-time. This has the potential to enable deaf users to access previously difficult-to-reach information. The use of automatic transcription applications can provide significant benefits for deaf users. Firstly, it allows pupils and students to follow conversations and presentations better, improving their listening skills. Second, users can participate more actively in various social and educational activities, feeling more confident in interacting with friends, colleagues and instructors. Apart from that, the use of this application can also help in the development of their spoken language, because students can view the transcribed text while listening to the conversation. However, there are several challenges that need to be overcome in implementing this automatic transcription application. These include accuracy of transcription, availability of appropriate hardware and software, and community understanding of the needs and rights of deaf users. Therefore, collaboration between technology developers, the deaf community, and other interested parties is important in ensuring that automated transcription applications can be used effectively to increase inclusivity for deaf users.

Data Flow Construction and Quality Evaluation of Electronic Source Data in Clinical Trials: Pilot Study Based on Hospital Electronic Medical Records in China.

The traditional clinical trial data collection process requires a clinical research coordinator who is authorized by the investigators to read from the hospital's electronic medical record. Using electronic source data opens a new path to extract patients' data from electronic health records (EHRs) and transfer them directly to an electronic data capture (EDC) system; this method is often referred to as eSource. eSource technology in a clinical trial data flow can improve data quality without compromising timeliness. At the same time, improved data collection efficiency reduces clinical trial costs. This study aims to explore how to extract clinical trial-related data from hospital EHR systems, transform the data into a format required by the EDC system, and transfer it into sponsors' environments, and to evaluate the transferred data sets to validate the availability, completeness, and accuracy of building an eSource dataflow. A prospective clinical trial study registered on the Drug Clinical Trial Registration and Information Disclosure Platform was selected, and the following data modules were extracted from the structured data of 4 case report forms: demographics, vital signs, local laboratory data, and concomitant medications. The extracted data was mapped and transformed, deidentified, and transferred to the sponsor's environment. Data validation was performed based on availability, completeness, and accuracy. In a secure and controlled data environment, clinical trial data was successfully transferred from a hospital EHR to the sponsor's environment with 100% transcriptional accuracy, but the availability and completeness of the data could be improved. Data availability was low due to some required fields in the EDC system not being available directly in the EHR. Some data is also still in an unstructured or paper-based format. The top-level design of the eSource technology and the construction of hospital electronic data standards should help lay a foundation for a full electronic data flow from EHRs to EDC systems in the future.

How ready is speech-to-text for psychological language research? Evaluating the validity of AI-generated English transcripts for analyzing free-spoken responses in younger and older adults.

For the longest time, the gold standard in preparing spoken language corpora for text analysis in psychology was using human transcription. However, such standard comes at extensive cost, and creates barriers to quantitative spoken language analysis that recent advances in speech-to-text technology could address. The current study quantifies the accuracy of AI-generated transcripts compared to human-corrected transcripts across younger (n = 100) and older (n = 92) adults and two spoken language tasks. Further, it evaluates the validity of Linguistic Inquiry and Word Count (LIWC)-features extracted from these two kinds of transcripts, as well as transcripts specifically prepared for LIWC analyses via tagging. We find that overall, AI-generated transcripts are highly accurate with a word error rate of 2.50% to 3.36%, albeit being slightly less accurate for younger compared to older adults. LIWC features extracted from either transcripts are highly correlated, while the tagging procedure significantly alters filler word categories. Based on these results, automatic speech-to-text appears to be ready for psychological language research when using spoken language tasks in relatively quiet environments, unless filler words are of interest to researchers.

Comparing perception of L1 and L2 English by human listeners and machines: Effect of interlocutor adaptationsa).

Speakers tailor their speech to different types of interlocutors. For example, speech directed to voice technology has different acoustic-phonetic characteristics than speech directed to a human. The present study investigates the perceptual consequences of human- and device-directed registers in English. We compare two groups of speakers: participants whose first language is English (L1) and bilingual L1 Mandarin-L2 English talkers. Participants produced short sentences in several conditions: an initial production and a repeat production after a human or device guise indicated either understanding or misunderstanding. In experiment 1, a separate group of L1 English listeners heard these sentences and transcribed the target words. In experiment 2, the same productions were transcribed by an automatic speech recognition (ASR) system. Results show that transcription accuracy was highest for L1 talkers for both human and ASR transcribers. Furthermore, there were no overall differences in transcription accuracy between human- and device-directed speech. Finally, while human listeners showed an intelligibility benefit for coda repair productions, the ASR transcriber did not benefit from these enhancements. Findings are discussed in terms of models of register adaptation, phonetic variation, and human-computer interaction.

SPEECH-TO-TEXT TRANSCRIPTION FOR CORPORATION GRIEVANCE CALLS

To evaluate and enhance an existing open-source speech-to-text transcription tool for accurately converting feedback calls about citizen grievances into English text. The focus lies on achieving measurable improvements in transcription accuracy across calls made in Tamil or English. Rather than creating a new system, the project concentrates on refining an established open-source solution. The endeavor spans domains including Natural Language Processing (NLP), speech recognition algorithms, and machine learning. Through this effort, the objective is to enhance the tool's efficacy in processing multilingual data and better serve the needs of diverse linguistic communities in citizen grievances.