RESUMEN

The importance of simulating patient behavior for medical assessment training has grown in recent decades due to the increasing variety of simulation tools, including standardized/simulated patients, humanoid and android robot-patients. Yet, there is still a need for improvement of current android robot-patients to accurately simulate patient behavior, among which taking into account their hearing loss is of particular importance. This paper is the first to consider hearing loss simulation in an android robot-patient and its results provide valuable insights for future developments. For this purpose, an open-source dataset of audio data and audiograms from human listeners was used to simulate the effect of hearing loss on an automatic speech recognition (ASR) system. The performance of the system was evaluated in terms of both word error rate (WER) and word information preserved (WIP). Comparing different ASR models commonly used in robotics, it appears that the model size alone is insufficient to predict ASR performance in presence of simulated hearing loss. However, though absolute values of WER and WIP do not predict the intelligibility for human listeners, they do highly correlate with it and thus could be used, for example, to compare the performance of hearing aid algorithms.

RESUMEN

Given an orthographic transcription, forced alignment systems automatically determine boundaries between segments in speech, facilitating the use of large corpora. In the present paper, we introduce a neural network-based forced alignment system, the Mason-Alberta Phonetic Segmenter (MAPS). MAPS serves as a testbed for two possible improvements we pursue for forced alignment systems. The first is treating the acoustic model as a tagger, rather than a classifier, motivated by the common understanding that segments are not truly discrete and often overlap. The second is an interpolation technique to allow more precise boundaries than the typical 10 ms limit in modern systems. During testing, all system configurations we trained significantly outperformed the state-of-the-art Montreal Forced Aligner in the 10 ms boundary placement tolerance threshold. The greatest difference achieved was a 28.13 % relative performance increase. The Montreal Forced Aligner began to slightly outperform our models at around a 30 ms tolerance. We also reflect on the training process for acoustic modeling in forced alignment, highlighting how the output targets for these models do not match phoneticians' conception of similarity between phones and that reconciling this tension may require rethinking the task and output targets or how speech itself should be segmented.

RESUMEN

Background: Speech analysis has been expected to help as a screening tool for early detection of Alzheimer's disease (AD) and mild-cognitively impairment (MCI). Acoustic features and linguistic features are usually used in speech analysis. However, no studies have yet determined which type of features provides better screening effectiveness, especially in the large aging population of China. Objective: Firstly, to compare the screening effectiveness of acoustic features, linguistic features, and their combination using the same dataset. Secondly, to develop Chinese automated diagnosis model using self-collected natural discourse data obtained from native Chinese speakers. Methods: A total of 92 participants from communities in Shanghai, completed MoCA-B and a picture description task based on the Cookie Theft under the guidance of trained operators, and were divided into three groups including AD, MCI, and heathy control (HC) based on their MoCA-B score. Acoustic features (Pitches, Jitter, Shimmer, MFCCs, Formants) and linguistic features (part-of-speech, type-token ratio, information words, information units) are extracted. The machine algorithms used in this study included logistic regression, random forest (RF), support vector machines (SVM), Gaussian Naive Bayesian (GNB), and k-Nearest neighbor (kNN). The validation accuracies of the same ML model using acoustic features, linguistic features, and their combination were compared. Results: The accuracy with linguistic features is generally higher than acoustic features in training. The highest accuracy to differentiate HC and AD is 80.77% achieved by SVM, based on all the features extracted from the speech data, while the highest accuracy to differentiate HC and AD or MCI is 80.43% achieved by RF, based only on linguistic features. Conclusion: Our results suggest the utility and validity of linguistic features in the automated diagnosis of cognitive impairment, and validated the applicability of automated diagnosis for Chinese language data.

Asunto(s)

RESUMEN

This study presents a model of automatic speech recognition (ASR) that is designed to diagnose pronunciation issues in children with speech sound disorders (SSDs) to replace manual transcriptions in clinical procedures. Because ASR models trained for general purposes mainly predict input speech into standard spelling words, well-known high-performance ASR models are not suitable for evaluating pronunciation in children with SSDs. We fine-tuned the wav2vec2.0 XLS-R model to recognise words as they are pronounced by children, rather than converting the speech into their standard spelling words. The model was fine-tuned with a speech dataset of 137 children with SSDs pronouncing 73 Korean words that are selected for actual clinical diagnosis. The model's Phoneme Error Rate (PER) was only 10% when its predictions of children's pronunciations were compared to human annotations of pronunciations as heard. In contrast, despite its robust performance on general tasks, the state-of-the-art ASR model Whisper showed limitations in recognising the speech of children with SSDs, with a PER of approximately 50%. While the model still requires improvement in terms of the recognition of unclear pronunciation, this study demonstrates that ASR models can streamline complex pronunciation error diagnostic procedures in clinical fields.

RESUMEN

In air traffic control (ATC), speech communication with radio transmission is the primary way to exchange information between the controller and the pilot. As a result, the integration of automatic speech recognition (ASR) systems holds immense potential for reducing controllers' workload and plays a crucial role in various ATC scenarios, which is particularly significant for ATC research. This article provides a comprehensive review of ASR technology's applications in the ATC communication system. Firstly, it offers a comprehensive overview of current research, including ATC corpora, ASR models, evaluation measures and application scenarios. A more comprehensive and accurate evaluation methodology tailored for ATC is proposed, considering advancements in communication sensing systems and deep learning techniques. This methodology helps researchers in enhancing ASR systems and improving the overall performance of ATC systems. Finally, future research recommendations are identified based on the primary challenges and issues. The authors sincerely hope this work will serve as a clear technical roadmap for ASR endeavors within the ATC domain and make a valuable contribution to the research community.

RESUMEN

The Web has become an essential resource but is not yet accessible to everyone. Assistive technologies and innovative, intelligent frameworks, for example, those using conversational AI, help overcome some exclusions. However, some users still experience barriers. This paper shows how a human-centered approach can shed light on technology limitations and gaps. It reports on a three-step process (focus group, co-design, and preliminary validation) that we adopted to investigate how people with speech impairments, e.g., dysarthria, browse the Web and how barriers can be reduced. The methodology helped us identify challenges and create new solutions, i.e., patterns for Web browsing, by combining voice-based conversational AI, customized for impaired speech, with techniques for the visual augmentation of web pages. While current trends in AI research focus on more and more powerful large models, participants remarked how current conversational systems do not meet their needs, and how it is important to consider each one's specificity for a technology to be called inclusive.

Asunto(s)

RESUMEN

This study introduces a specialized Automatic Speech Recognition (ASR) system, leveraging the Whisper Large-v2 model, specifically adapted for radiological applications in the French language. The methodology focused on adapting the model to accurately transcribe medical terminology and diverse accents within the French language context, achieving a notable Word Error Rate (WER) of 17.121%. This research involved extensive data collection and preprocessing, utilizing a wide range of French medical audio content. The results demonstrate the system's effectiveness in transcribing complex radiological data, underscoring its potential to enhance medical documentation efficiency in French-speaking clinical settings. The discussion extends to the broader implications of this technology in healthcare, including its potential integration with electronic health records (EHRs) and its utility in medical education. This study also explores future research directions, such as tailoring ASR systems to specific medical specialties and languages. Overall, this research contributes significantly to the field of medical ASR systems, presenting a robust tool for radiological transcription in the French language and paving the way for advanced technology-enhanced healthcare solutions.

RESUMEN

This research presents the development of a cutting-edge real-time multilingual speech recognition and speaker diarization system that leverages OpenAI's Whisper model. The system specifically addresses the challenges of automatic speech recognition (ASR) and speaker diarization (SD) in dynamic, multispeaker environments, with a focus on accurately processing Mandarin speech with Taiwanese accents and managing frequent speaker switches. Traditional speech recognition systems often fall short in such complex multilingual and multispeaker contexts, particularly in SD. This study, therefore, integrates advanced speech recognition with speaker diarization techniques optimized for real-time applications. These optimizations include handling model outputs efficiently and incorporating speaker embedding technology. The system was evaluated using data from Taiwanese talk shows and political commentary programs, featuring 46 diverse speakers. The results showed a promising word diarization error rate (WDER) of 2.68% in two-speaker scenarios and 11.65% in three-speaker scenarios, with an overall WDER of 6.96%. This performance is comparable to that of non-real-time baseline models, highlighting the system's ability to adapt to various complex conversational dynamics, a significant advancement in the field of real-time multilingual speech processing.

RESUMEN

Background: In today's world, numerous applications integral to various facets of daily life include automatic speech recognition methods. Thus, the development of a successful automatic speech recognition system can significantly augment the convenience of people's daily routines. While many automatic speech recognition systems have been established for widely spoken languages like English, there has been insufficient progress in developing such systems for less common languages such as Turkish. Moreover, due to its agglutinative structure, designing a speech recognition system for Turkish presents greater challenges compared to other language groups. Therefore, our study focused on proposing deep learning models for automatic speech recognition in Turkish, complemented by the integration of a language model. Methods: In our study, deep learning models were formulated by incorporating convolutional neural networks, gated recurrent units, long short-term memories, and transformer layers. The Zemberek library was employed to craft the language model to improve system performance. Furthermore, the Bayesian optimization method was applied to fine-tune the hyper-parameters of the deep learning models. To evaluate the model's performance, standard metrics widely used in automatic speech recognition systems, specifically word error rate and character error rate scores, were employed. Results: Upon reviewing the experimental results, it becomes evident that when optimal hyper-parameters are applied to models developed with various layers, the scores are as follows: Without the use of a language model, the Turkish Microphone Speech Corpus dataset yields scores of 22.2 -word error rate and 14.05-character error rate, while the Turkish Speech Corpus dataset results in scores of 11.5 -word error rate and 4.15 character error rate. Upon incorporating the language model, notable improvements were observed. Specifically, for the Turkish Microphone Speech Corpus dataset, the word error rate score decreased to 9.85, and the character error rate score lowered to 5.35. Similarly, the word error rate score improved to 8.4, and the character error rate score decreased to 2.7 for the Turkish Speech Corpus dataset. These results demonstrate that our model outperforms the studies found in the existing literature.

RESUMEN

Introduction: Automatic recognition of stutters (ARS) from speech recordings can facilitate objective assessment and intervention for people who stutter. However, the performance of ARS systems may depend on how the speech data are segmented and labelled for training and testing. This study compared two segmentation methods: event-based, which delimits speech segments by their fluency status, and interval-based, which uses fixed-length segments regardless of fluency. Methods: Machine learning models were trained and evaluated on interval-based and event-based stuttered speech corpora. The models used acoustic and linguistic features extracted from the speech signal and the transcriptions generated by a state-of-the-art automatic speech recognition system. Results: The results showed that event-based segmentation led to better ARS performance than interval-based segmentation, as measured by the area under the curve (AUC) of the receiver operating characteristic. The results suggest differences in the quality and quantity of the data because of segmentation method. The inclusion of linguistic features improved the detection of whole-word repetitions, but not other types of stutters. Discussion: The findings suggest that event-based segmentation is more suitable for ARS than interval-based segmentation, as it preserves the exact boundaries and types of stutters. The linguistic features provide useful information for separating supra-lexical disfluencies from fluent speech but may not capture the acoustic characteristics of stutters. Future work should explore more robust and diverse features, as well as larger and more representative datasets, for developing effective ARS systems.

RESUMEN

A practical speech audiometry tool is the digits-in-noise (DIN) test for hearing screening of populations of varying ages and hearing status. The test is usually conducted by a human supervisor (e.g., clinician), who scores the responses spoken by the listener, or online, where software scores the responses entered by the listener. The test has 24-digit triplets presented in an adaptive staircase procedure, resulting in a speech reception threshold (SRT). We propose an alternative automated DIN test setup that can evaluate spoken responses whilst conducted without a human supervisor, using the open-source automatic speech recognition toolkit, Kaldi-NL. Thirty self-reported normal-hearing Dutch adults (19-64 years) completed one DIN + Kaldi-NL test. Their spoken responses were recorded and used for evaluating the transcript of decoded responses by Kaldi-NL. Study 1 evaluated the Kaldi-NL performance through its word error rate (WER), percentage of summed decoding errors regarding only digits found in the transcript compared to the total number of digits present in the spoken responses. Average WER across participants was 5.0% (range 0-48%, SD = 8.8%), with average decoding errors in three triplets per participant. Study 2 analyzed the effect that triplets with decoding errors from Kaldi-NL had on the DIN test output (SRT), using bootstrapping simulations. Previous research indicated 0.70 dB as the typical within-subject SRT variability for normal-hearing adults. Study 2 showed that up to four triplets with decoding errors produce SRT variations within this range, suggesting that our proposed setup could be feasible for clinical applications.

Asunto(s)

RESUMEN

The Voiceitt app is designed for people with dysarthric speech, to support vocal communication and access to voice-driven technologies. Sixty-six participants were recruited to test the Voiceitt app and share feedback with developers. Most had physical, sensory, or cognitive impairments in addition to atypical speech. The project team liaised with individuals, their families and local support teams to provide access to the app and associated equipment. Testing was user-led, with participants asked to identify and test use cases most relevant to their daily lives over three months or more. Ongoing technical support and training were provided remotely and in-person throughout their testing. Semi-structured interviews were used to collect feedback on users' experiences, with delivery adapted to individuals' needs and preferences. Informal feedback was collected through ongoing contact between participants, their families and support teams and the project team. User feedback has led to improvements to the user interface and functionality, including faster voice training, simplified navigation, the introduction of game-style features and of switch access as an alternative to touchscreen access. This work offers a case-study in meaningful engagement with diverse disabled users of assistive technology in commercial software development.

RESUMEN

BACKGROUND: Prior research has associated spoken language use with depression, yet studies often involve small or non-clinical samples and face challenges in the manual transcription of speech. This paper aimed to automatically identify depression-related topics in speech recordings collected from clinical samples. METHODS: The data included 3919 English free-response speech recordings collected via smartphones from 265 participants with a depression history. We transcribed speech recordings via automatic speech recognition (Whisper tool, OpenAI) and identified principal topics from transcriptions using a deep learning topic model (BERTopic). To identify depression risk topics and understand the context, we compared participants' depression severity and behavioral (extracted from wearable devices) and linguistic (extracted from transcribed texts) characteristics across identified topics. RESULTS: From the 29 topics identified, we identified 6 risk topics for depression: 'No Expectations', 'Sleep', 'Mental Therapy', 'Haircut', 'Studying', and 'Coursework'. Participants mentioning depression risk topics exhibited higher sleep variability, later sleep onset, and fewer daily steps and used fewer words, more negative language, and fewer leisure-related words in their speech recordings. LIMITATIONS: Our findings were derived from a depressed cohort with a specific speech task, potentially limiting the generalizability to non-clinical populations or other speech tasks. Additionally, some topics had small sample sizes, necessitating further validation in larger datasets. CONCLUSION: This study demonstrates that specific speech topics can indicate depression severity. The employed data-driven workflow provides a practical approach for analyzing large-scale speech data collected from real-world settings.

Asunto(s)

RESUMEN

Introduction: Objectively predicting speech intelligibility is important in both telecommunication and human-machine interaction systems. The classic method relies on signal-to-noise ratios (SNR) to successfully predict speech intelligibility. One exception is clear speech, in which a talker intentionally articulates as if speaking to someone who has hearing loss or is from a different language background. As a result, at the same SNR, clear speech produces higher intelligibility than conversational speech. Despite numerous efforts, no objective metric can successfully predict the clear speech benefit at the sentence level. Methods: We proposed a Syllable-Rate-Adjusted-Modulation (SRAM) index to predict the intelligibility of clear and conversational speech. The SRAM used as short as 1 s speech and estimated its modulation power above the syllable rate. We compared SRAM with three reference metrics: envelope-regression-based speech transmission index (ER-STI), hearing-aid speech perception index version 2 (HASPI-v2) and short-time objective intelligibility (STOI), and five automatic speech recognition systems: Amazon Transcribe, Microsoft Azure Speech-To-Text, Google Speech-To-Text, wav2vec2 and Whisper. Results: SRAM outperformed the three reference metrics (ER-STI, HASPI-v2 and STOI) and the five automatic speech recognition systems. Additionally, we demonstrated the important role of syllable rate in predicting speech intelligibility by comparing SRAM with the total modulation power (TMP) that was not adjusted by the syllable rate. Discussion: SRAM can potentially help understand the characteristics of clear speech, screen speech materials with high intelligibility, and convert conversational speech into clear speech.

RESUMEN

INTRODUCTION: Early detection of Alzheimer's disease and cognitive impairment is critical to improving the healthcare trajectories of aging adults, enabling early intervention and potential prevention of decline. METHODS: To evaluate multi-modal feature sets for assessing memory and cognitive impairment, feature selection and subsequent logistic regressions were used to identify the most salient features in classifying Rey Auditory Verbal Learning Test-determined memory impairment. RESULTS: Multimodal models incorporating graphomotor, memory, and speech and voice features provided the stronger classification performance (area under the curve = 0.83; sensitivity = 0.81, specificity = 0.80). Multimodal models were superior to all other single modality and demographics models. DISCUSSION: The current research contributes to the prevailing multimodal profile of those with cognitive impairment, suggesting that it is associated with slower speech with a particular effect on the duration, frequency, and percentage of pauses compared to normal healthy speech.

RESUMEN

In this work we document the development of an ASR system for the transcription of conversations between patient and doctor and we will point out the critical aspects of the domain. The system was trained with an acoustic base of spontaneous speech that has a domain language model and a supervised phonetic dictionary. Its performance was compared with two systems: a) NeMo End-to-End Conformers in Spanish and b) Google API ASR (Automatic Speech Recognition) Cloud. The evaluation was carried out on a set of 208 teleconsultations recorded during the year 2020. The WER (Word Error Rate) was evaluated in ASR, and Recall and F1 for recognized medical entities. In conclusion, the developed system performed better, reaching 72.5% accuracy in the domain of teleconsultations and an F1 for entity recognition of 0.80.

Asunto(s)

RESUMEN

OBJECTIVES: We aimed to investigate how errors from automatic speech recognition (ASR) systems affect dementia classification accuracy, specifically in the "Cookie Theft" picture description task. We aimed to assess whether imperfect ASR-generated transcripts could provide valuable information for distinguishing between language samples from cognitively healthy individuals and those with Alzheimer's disease (AD). METHODS: We conducted experiments using various ASR models, refining their transcripts with post-editing techniques. Both these imperfect ASR transcripts and manually transcribed ones were used as inputs for the downstream dementia classification. We conducted comprehensive error analysis to compare model performance and assess ASR-generated transcript effectiveness in dementia classification. RESULTS: Imperfect ASR-generated transcripts surprisingly outperformed manual transcription for distinguishing between individuals with AD and those without in the "Cookie Theft" task. These ASR-based models surpassed the previous state-of-the-art approach, indicating that ASR errors may contain valuable cues related to dementia. The synergy between ASR and classification models improved overall accuracy in dementia classification. CONCLUSION: Imperfect ASR transcripts effectively capture linguistic anomalies linked to dementia, improving accuracy in classification tasks. This synergy between ASR and classification models underscores ASR's potential as a valuable tool in assessing cognitive impairment and related clinical applications.

Asunto(s)

RESUMEN

Dysarthria is a common manifestation across cerebellar ataxias leading to impairments in communication, reduced social connections, and decreased quality of life. While dysarthria symptoms may be present in other neurological conditions, ataxic dysarthria is a perceptually distinct motor speech disorder, with the most prominent characteristics being articulation and prosody abnormalities along with distorted vowels. We hypothesized that uncertainty of vowel predictions by an automatic speech recognition system can capture speech changes present in cerebellar ataxia. Speech of participants with ataxia (N=61) and healthy controls (N=25) was recorded during the "picture description" task. Additionally, participants' dysarthric speech and ataxia severity were assessed on a Brief Ataxia Rating Scale (BARS). Eight participants with ataxia had speech and BARS data at two timepoints. A neural network trained for phoneme prediction was applied to speech recordings. Average entropy of vowel tokens predictions (AVE) was computed for each participant's recording, together with mean pitch and intensity standard deviations (MPSD and MISD) in the vowel segments. AVE and MISD demonstrated associations with BARS speech score (Spearman's rho=0.45 and 0.51), and AVE demonstrated associations with BARS total (rho=0.39). In the longitudinal cohort, Wilcoxon pairwise signed rank test demonstrated an increase in BARS total and AVE, while BARS speech and acoustic measures did not significantly increase. Relationship of AVE to both BARS speech and BARS total, as well as the ability to capture disease progression even in absence of measured speech decline, indicates the potential of AVE as a digital biomarker for cerebellar ataxia.

Asunto(s)

RESUMEN

Automatic speech recognition (ASR) has been an active area of research. Training with large annotated datasets is the key to the development of robust ASR systems. However, most available datasets are focused on high-resource languages like English, leaving a significant gap for low-resource languages. Among these languages is Punjabi, despite its large number of speakers, Punjabi lacks high-quality annotated datasets for accurate speech recognition. To address this gap, we introduce three labeled Punjabi speech datasets: Punjabi Speech (real speech dataset) and Google-synth/CMU-synth (synthesized speech datasets). The Punjabi Speech dataset consists of read speech recordings captured in various environments, including both studio and open settings. In addition, the Google-synth dataset is synthesized using Google's Punjabi text-to-speech cloud services. Furthermore, the CMU-synth dataset is created using the Clustergen model available in the Festival speech synthesis system developed by CMU. These datasets aim to facilitate the development of accurate Punjabi speech recognition systems, bridging the resource gap for this important language.

RESUMEN

Large pre-trained Automatic Speech Recognition (ASR) models have shown improved performance in low-resource languages due to the increased availability of benchmark corpora and the advantages of transfer learning. However, only a limited number of languages possess ample resources to fully leverage transfer learning. In such contexts, benchmark corpora become crucial for advancing methods. In this article, we introduce two new benchmark corpora designed for low-resource languages spoken in the Democratic Republic of the Congo: the Lingala Read Speech Corpus, with 4 h of labelled audio, and the Congolese Speech Radio Corpus, which offers 741 h of unlabelled audio spanning four significant low-resource languages of the region. During data collection, Lingala Read Speech recordings of thirty-two distinct adult speakers, each with a unique context under various settings with different accents, were recorded. Concurrently, Congolese Speech Radio raw data were taken from the archive of broadcast station, followed by a designed curation process. During data preparation, numerous strategies have been utilised for pre-processing the data. The datasets, which have been made freely accessible to all researchers, serve as a valuable resource for not only investigating and developing monolingual methods and approaches that employ linguistically distant languages but also multilingual approaches with linguistically similar languages. Using techniques such as supervised learning and self-supervised learning, they are able to develop inaugural benchmarking of speech recognition systems for Lingala and mark the first instance of a multilingual model tailored for four Congolese languages spoken by an aggregated population of 95 million. Moreover, two models were applied to this dataset. The first is supervised learning modelling and the second is for self-supervised pre-training.