Speech Recognition Data Research Articles

The application of WeChat navigation assistant

The field of artificial intelligence and deep learning has made significant progress in recent years, with the development of innovative techniques and technologies that have the potential to improve various aspects of human life. However, the elderly population faces significant challenges when it comes to utilizing electronic devices, as the demand for convenient and easy-to-use products in their daily lives remains unmet. To address this challenge, this paper proposes the design of a software solution, consisting of four key components. Firstly, the software application is simulated, and the process of software operation when the elderly use WeChat is modeled. Secondly, automatic speech recognition technology is utilized to enhance the usability of WeChat for the elderly. Thirdly, a dataset is created for data collection, processing, and analysis, with the aim of constantly improving the software. Finally, evaluation methods such as formative and summative evaluations are utilized to assess and enhance the effectiveness of the software. The proposed software solution has the potential to significantly improve the quality of life of the elderly population by enabling them to better access and utilize electronic devices. Moreover, the incorporation of automatic speech recognition technology and data analysis for continuous improvement has the potential to contribute to the advancement of the field of artificial intelligence and deep learning. Further research should focus on refining the software solution to better cater to the specific needs of the elderly population.

Reconsidering Read and Spontaneous Speech: Causal Perspectives on the Generation of Training Data for Automatic Speech Recognition

Superficially, read and spontaneous speech—the two main kinds of training data for automatic speech recognition—appear as complementary, but are equal: pairs of texts and acoustic signals. Yet, spontaneous speech is typically harder for recognition. This is usually explained by different kinds of variation and noise, but there is a more fundamental deviation at play: for read speech, the audio signal is produced by recitation of the given text, whereas in spontaneous speech, the text is transcribed from a given signal. In this review, we embrace this difference by presenting a first introduction of causal reasoning into automatic speech recognition, and describing causality as a tool to study speaking styles and training data. After breaking down the data generation processes of read and spontaneous speech and analysing the domain from a causal perspective, we highlight how data generation by annotation must affect the interpretation of inference and performance. Our work discusses how various results from the causality literature regarding the impact of the direction of data generation mechanisms on learning and prediction apply to speech data. Finally, we argue how a causal perspective can support the understanding of models in speech processing regarding their behaviour, capabilities, and limitations.

An online intelligent electronic medical record system via speech recognition

Traditional electronic medical record systems in hospitals rely on healthcare workers to manually enter patient information, resulting in healthcare workers having to spend a significant amount of time each day filling out electronic medical records. This inefficient interaction seriously affects the communication between doctors and patients and reduces the speed at which doctors can diagnose patients’ conditions. The rapid development of deep learning–based speech recognition technology promises to improve this situation. In this work, we build an online electronic medical record system based on speech interaction. The system integrates a medical linguistic knowledge base, a specialized language model, a personalized acoustic model, and a fault-tolerance mechanism. Hence, we propose and develop an advanced electronic medical record system approach with multi-accent adaptive technology for avoiding the mistakes caused by accents, and it improves the accuracy of speech recognition obviously. For testing the proposed speech recognition electronic medical record system, we construct medical speech recognition data sets using audio and electronic medical records from real medical environments. On the data sets from real clinical scenarios, our proposed algorithm significantly outperforms other machine learning algorithms. Furthermore, compared to traditional electronic medical record systems that rely on keyboard inputs, our system is much more efficient, and its accuracy rate increases with the increasing online time of the proposed system. Our results show that the proposed electronic medical record system is expected to revolutionize the traditional working approach of clinical departments, and it serves more efficient in clinics with low time consumption compared with traditional electronic medical record systems depending on keyboard inputs, which has less recording mistakes and lows down the time consumption in modification of medical recordings; due to the proposed speech recognition electronic medical record system is built on knowledge database of medical terms, so it has a good generalized application and adaption in the clinical scenarios for hospitals.

Maturation of Speech-in-Speech Recognition for Whispered and Voiced Speech.

Some speech recognition data suggest that children rely less on voice pitch and harmonicity to support auditory scene analysis than adults. Two experiments evaluated development of speech-in-speech recognition using voiced speech and whispered speech, which lacks the harmonic structure of voiced speech. Listeners were 5- to 7-year-olds and adults with normal hearing. Targets were monosyllabic words organized into three-word sets that differ in vowel content. Maskers were two-talker or one-talker streams of speech. Targets and maskers were recorded by different female talkers in both voiced and whispered speaking styles. For each masker, speech reception thresholds (SRTs) were measured in all four combinations of target and masker speech, including matched and mismatched speaking styles for the target and masker. Children performed more poorly than adults overall. For the two-talker masker, this age effect was smaller for the whispered target and masker than for the other three conditions. Children's SRTs in this condition were predominantly positive, suggesting that they may have relied on a wholistic listening strategy rather than segregating the target from the masker. For the one-talker masker, age effects were consistent across the four conditions. Reduced informational masking for the one-talker masker could be responsible for differences in age effects for the two maskers. A benefit of mismatching the target and masker speaking style was observed for both target styles in the two-talker masker and for the voiced targets in the one-talker masker. These results provide no compelling evidence that young school-age children and adults are differentially sensitive to the cues present in voiced and whispered speech. Both groups benefit from mismatches in speaking style under some conditions. These benefits could be due to a combination of reduced perceptual similarity, harmonic cancelation, and differences in energetic masking.

Intelligent Robot English Speech Recognition Method Based on Online Database

In order to solve the problem of low accuracy of traditional English speech recognition, an intelligent robot English speech recognition method based on online database is proposed. A speech recognition device is installed on the intelligent robot as the hardware support for running the speech recognition method. The online English speech standard database is constructed to provide reference data for speech recognition. The real-time speech information is collected, and the speech signal is preprocessed by pre-emphasis, framing, windowing and other steps. According to the principle of speech signal generation, the features of speech signal are extracted, and the results of English speech recognition are obtained by similarity calculation and matching. Compared with the traditional recognition method, the experimental results show that the recognition rate of the optimised speech recognition method is improved by 1.3%, i.e. the recognition accuracy is improved.

Voice Conversion Based Augmentation and a Hybrid CNN-LSTM Model for Improving Speaker-Independent Keyword Recognition on Limited Datasets

Keyword recognition is the basis of speech recognition, and its application is rapidly increasing in keyword spotting, robotics, and smart home surveillance. Because of these advanced applications, improving the accuracy of keyword recognition is crucial. In this paper, we proposed voice conversion (VC) - based augmentation to increase the limited training dataset and a fusion of a convolutional neural network (CNN) and long-short term memory (LSTM) model for robust speaker-independent isolated keyword recognition. Collecting and preparing a sufficient amount of voice data for speaker-independent speech recognition is a tedious and bulky task. To overcome this, we generated new raw voices from the original voices using an auxiliary classifier conditional variational autoencoder (ACVAE) method. In this study, the main intention of voice conversion is to obtain numerous and various human-like keywords’ voices that are not identical to the source and target speakers’ pronunciation. Parallel VC was used to accurately maintain the linguistic content. We examined the performance of the proposed voice conversion augmentation techniques using robust deep neural network algorithms. Original training data, excluding generated voice using other data augmentation and regularization techniques, were considered as the baseline. The results showed that incorporating voice conversion augmentation into the baseline augmentation techniques and applying the CNN-LSTM model improved the accuracy of isolated keyword recognition.

Masked Speech Recognition by Normal-Hearing 6-13-Year-Olds in Conditions With and Without Interaural Difference Cues.

Using the Danish 'børneDAT' corpus, the current study aimed to (1) collect normative masked speech recognition data for 6-13-year-olds in conditions with and without interaural difference cues, (2) evaluate the test-retest reliability of these measurements, and (3) compare two widely used measures of binaural/spatial benefit in terms of the obtained scores. Seventy-four children and 17 young adults with normal hearing participated. Using headphone presentation, speech recognition thresholds (SRTs) were measured twice at two separate visits in four conditions. In the first two conditions, børneDAT sentences were presented in diotic stationary speech-shaped noise, with the sentences either interaurally in-phase ('N0S0') or interaurally out-of-phase ('N0S180'). In the other two conditions, børneDAT sentences were simulated to come from 0° azimuth and two running speech maskers from either 0° azimuth ('co-located') or ±90° azimuth ('spatially separated'). In relative terms, the children achieved lower SRTs in stationary noise than in competing speech, whereas the adults showed the opposite pattern. 12-13-year-old children achieved adult-like performance in all but the co-located condition. Younger children showed generally immature speech recognition abilities. Test-retest reliability was highest for the SRTs in stationary noise and lowest for the spatial benefit scores. Mean benefit was comparable for the two measures and participant groups, and the two sets of scores were not correlated with each other. Developmental effects were most pronounced in the conditions with interaural difference cues. In conclusion, reference data for the børneDAT corpus obtained under different acoustic conditions are available that can guide future research and potential clinical applications.

An enhanced user’s speech recognition data analysis and expectation in intelligent personal assistant

Intelligent Personal Assistant (IPA) adoption has ushered in a technological revolution in today's society. There are several well-known IPA agents, like Google Assistant and Cortana, which were both created by Microsoft. The benefits of IPAs' speech comprehension capabilities have resulted in an increase in IPA users year after year. The goal of the study is to look at how an IPA's speech comprehension capability could be improved in the future, making it easier for IPA users to handle their everyday routines. This study will employ a survey and an unstructured interview to learn about users' expectations for an IPA's speech comprehension capabilities and how IPA adaptation affects their daily routine management. To fulfill the research's goal, an IPA with a daily routine management system will be suggested using the methodology and project design that have been determined.

"Syllabic Units Automatically Segmented Data for Continuous Speech Recognition"

We present novel approach for constant speech processing in which the detection and recognition tasks are separated A syllable is utilized as a measure both to detection and localization. A minimal phase’s group delay characteristic approach and an utterance isolated style are used to segment the speech signal at the boundaries of syllabic units. For two Indigenous languages, an HMM recognizing system has been created. Viterbi algorithm-based methods are suggested to solve recognition problems caused by shifts in segment borders and syllabic unit merging.

Robust Speech Recognition Model Using Multi-Source Federal Learning after Distillation and Deep Edge Intelligence

In order to further explore and solve the problem of distributed training edge intelligent application in edge devices close to the data side, a joint training method of multi-source federated learning is proposed for the robust speech recognition model. The joint training is performed directly on the edge computing device close to the data supply side. Under the open source Chinese speech recognition data set, the recognition accuracy is 95%. Compared with federated learning, there is only 2% performance reduction, but the communication load is reduced by 90%.

Remote collection of speech recognition data: Considerations for testing children

This talk will provide an overview of the methods and lessons learned over the course of three studies in which children’s masked speech recognition performance was assessed remotely. The first study evaluated the binaural intelligibility level difference in typically developing children using families’ personal computers and headphones. The second study examined the impact of face coverings on consonant identification in noise with and without access to visual cues. Participants were children with and without hearing loss. The third experiment evaluated the feasibility of remote speech recognition testing for children with Down syndrome. The latter two experiments entailed providing families with laboratory hardware, following appropriate social distancing and infection control measures. Findings from all three studies will be discussed, highlighting comparisons with data obtained in the laboratory. Recommendations for remote studies involving children will be provided, based on our experiences and the resultant data.

Transfer learning and SpecAugment applied to SSVEP based BCI classification

ObjectiveWe used deep convolutional neural networks (DCNNs) to classify electroencephalography (EEG) signals in a steady-state visually evoked potentials (SSVEP) based single-channel brain–computer interface (BCI), which does not require calibration on the user. MethodsEEG signals were converted to spectrograms and served as input to train DCNNs using the transfer learning technique. We also modified and applied a data augmentation method, SpecAugment, generally employed for speech recognition. Furthermore, for comparison purposes, we classified the SSVEP dataset using support-vector machines (SVMs) and filter bank canonical correlation analysis (FBCCA). ResultsExcluding the evaluated user's data from the fine-tuning process, we reached 82.2% mean test accuracy and 0.825 mean F1-Score on 35 subjects from an open dataset, using a small data length (0.5 s), only one electrode (Oz) and the DCNN with transfer learning, window slicing (WS) and SpecAugment's time masks. ConclusionThe DCNN results surpassed SVM and FBCCA performances, using a single electrode and a small data length. Transfer learning provided minimal accuracy change, but made training faster. SpecAugment created a small performance improvement and was successfully combined with WS, yielding higher accuracies. SignificanceWe present a new methodology to solve the problem of SSVEP classification using DCNNs. We also modified a speech recognition data augmentation technique and applied it to the context of BCIs. The presented methodology surpassed performances obtained with FBCCA and SVMs (more traditional SSVEP classification methods) in BCIs with small data lengths and one electrode. This type of BCI can be used to develop small and fast systems.

Extending automatic transcripts in a unified data representation towards a prosodic-based metadata annotation and evaluation

This paper describes a framework that extends automatic speech transcripts in order to accommodate relevant information coming from manual transcripts, the speech signal itself, and other resources, like lexica. The proposed framework automatically collects, relates, computes, and stores all relevant information together in a self-contained data source, making it possible to easily provide a wide range of interconnected information suitable for speech analysis, training, and evaluating a number of automatic speech processing tasks. The main goal of this framework is to integrate different linguistic and paralinguistic layers of knowledge for a more complete view of their representation and interactions in several domains and languages. The processing chain is composed of two main stages, where the first consists of integrating the relevant manual annotations in the speech recognition data, and the second consists of further enriching the previous output in order to accommodate prosodic information. The described framework has been used for the identification and analysis of structural metadata in automatic speech transcripts. Initially put to use for automatic detection of punctuation marks and for capitalization recovery from speech data, it has also been recently used for studying the characterization of disfluencies in speech. It was already applied to several domains of Portuguese corpora, and also to English and Spanish Broadcast News corpora.

Any-to-Many Voice Conversion With Location-Relative Sequence-to-Sequence Modeling

This paper proposes an any-to-many location-relative, sequence-to-sequence (seq2seq), non-parallel voice conversion approach, which utilizes text supervision during training. In this approach, we combine a bottle-neck feature extractor (BNE) with a seq2seq synthesis module. During the training stage, an encoder-decoder-based hybrid connectionist-temporal-classification-attention (CTC-attention) phoneme recognizer is trained, whose encoder has a bottle-neck layer. A BNE is obtained from the phoneme recognizer and is utilized to extract speaker-independent, dense and rich spoken content representations from spectral features. Then a multi-speaker location-relative attention based seq2seq synthesis model is trained to reconstruct spectral features from the bottle-neck features, conditioning on speaker representations for speaker identity control in the generated speech. To mitigate the difficulties of using seq2seq models to align long sequences, we down-sample the input spectral feature along the temporal dimension and equip the synthesis model with a discretized mixture of logistic (MoL) attention mechanism. Since the phoneme recognizer is trained with large speech recognition data corpus, the proposed approach can conduct any-to-many voice conversion. Objective and subjective evaluations show that the proposed any-to-many approach has superior voice conversion performance in terms of both naturalness and speaker similarity. Ablation studies are conducted to confirm the effectiveness of feature selection and model design strategies in the proposed approach. The proposed VC approach can readily be extended to support any-to-any VC (also known as one/few-shot VC), and achieve high performance according to objective and subjective evaluations.

Sound Classification Using Python

Sound assumes a significant part in human existence. It is one of the fundamental tangible data which we get or see from the climate and their components which have three principal credits viz. Sufficiency (Loudness of the sound), Frequency (The pitch of the sound), Timbre (Quality of the sound or the personality of the sound for example the Sound contrast between a piano and a violin). It is an event generated from the action. Humans are highly efficient to learn and recognize new and various types of sounds and sound events. There is a lot of research work going on Automatic sound classification and it is used in various real-world applications. The paper proposes an examination of an establishment disturbance classifier reliant upon a model affirmation approach using a neural organization. The signs submitted to the neural association are depicted through a lot of 12 MFCC (Mel Frequency Cepstral Coefficient) limits routinely present toward the front finish of an adaptable terminal. The introduction of the classifier, assessed as far as percent misclassification, show an exactness going between 73 % and 95 % relying upon the term of the choice window. Transmitting sound using a machine and expecting an output is considered a highly accurate deep learning task. This technology is used in our smartphones with mobile assistants such as Siri, Alexa, Google Assistant. In the case of the Google Speech recognition data set over 94 percent accuracy is obtained when trying to identify one of 20 words, silence or unknown. It is a very difficult task to recognize audio or sound events systematically and work on it for identification and give output. We are going to work on it using python programming language and some deep learning techniques. It’s a basic model that we are trying to develop, taking the next step to the innovative model that can help society and also which represent the innovative ideas of Engineering Students.

Template-based mapping of dynamic motifs in tissue morphogenesis.

Tissue morphogenesis relies on repeated use of dynamic behaviors at the levels of intracellular structures, individual cells, and cell groups. Rapidly accumulating live imaging datasets make it increasingly important to formalize and automate the task of mapping recurrent dynamic behaviors (motifs), as it is done in speech recognition and other data mining applications. Here, we present a “template-based search” approach for accurate mapping of sub- to multi-cellular morphogenetic motifs using a time series data mining framework. We formulated the task of motif mapping as a subsequence matching problem and solved it using dynamic time warping, while relying on high throughput graph-theoretic algorithms for efficient exploration of the search space. This formulation allows our algorithm to accurately identify the complete duration of each instance and automatically label different stages throughout its progress, such as cell cycle phases during cell division. To illustrate our approach, we mapped cell intercalations during germband extension in the early Drosophila embryo. Our framework enabled statistical analysis of intercalary cell behaviors in wild-type and mutant embryos, comparison of temporal dynamics in contracting and growing junctions in different genotypes, and the identification of a novel mode of iterative cell intercalation. Our formulation of tissue morphogenesis using time series opens new avenues for systematic decomposition of tissue morphogenesis.

Privacy-centered design for social robots

AbstractSocial robots as companions play an increasingly important role in our everyday life. However, reaching the full potential of social robots and the interaction between humans and robots requires permanent collection and processing of personal data of users, e.g. video and audio data for image and speech recognition. In order to foster user acceptance, trust and to address legal requirements as the General Data Protection Regulation of the EU, privacy needs to be integrated in the design process of social robots. The Privacy by Design approach by Cavoukian indicates the relevance of a privacy-respecting development and outlines seven abstract principle.In this paper two methods as a hands-on guideline to fulfill the principles are presented and discussed in the content of the Privacy by Design approach. Privacy risks of a typical robot scenario are identified, analyzed and solutions are proposed on the basis of theseven types of privacyand theprivacy protection goals.

Prelingual Deaf Children Treated With Cochlear Implant: Monitoring Performance With Percentiles.

To develop a percentile ranking system driven by speech recognition data obtained from different groups of patients treated with a cochlear implant to serve as a tool to monitor the progress of these patients. Prospective study. Tertiary referral center. Diagnosed with a bilateral, profound sensorineural hearing loss treated with a unilateral cochlear implant. Diagnostic. Six different percentiles (p) were classified taking into account the correlation between speech recognition outcome scores and age at implantation, with reference to the onset of hearing loss. Four hundred sixteen prelingual patients were included. These subjects were divided into subgroups depending on age at implantation.Prelingual group, from the fifth year after implantation, p50 centered on the following percentages of correct words in each subgroup: 100, 94.6, 91.4, 91.0, 79.2, and 63.1% in children implanted under 12 months, 1, 2, 3, 4 to 6, 7 to 10 years, respectively. After a 12-year follow-up, a significant negative correlation between age at implantation and speech recognition was observed in both prelinguals (Rhos=-0.578, p<0.001). A percentile system was developed to monitor the postimplant progress of prelingual deaf implanted patients, with potential applications in patient follow-up and handling circumstances that may deteriorate results.

ML-SLSTSVM: a new structural least square twin support vector machine for multi-label learning

Multi-label learning (MLL) is a special supervised learning task, where any single instance possibly belongs to several classes simultaneously. Nowadays, MLL methods are increasingly required by modern applications, such as protein function classification, speech recognition and textual data classification. In this paper, a structural least square twin support vector machine (SLSTSVM) classifier for multi-label learning is presented. This proposed ML-SLSTSVM focuses on the cluster-based structural information of the corresponding class in each optimization problem, which is vital for designing a good classifier in different real-world problems. This method is extended to a nonlinear version by the kernel trick. Experimental results demonstrate that proposed method is superior in generalization performance to other classifiers.

Speech Perception in Complex Acoustic Environments: Developmental Effects.

The ability to hear and understand speech in complex acoustic environments follows a prolonged time course of development. The purpose of this article is to provide a general overview of the literature describing age effects in susceptibility to auditory masking in the context of speech recognition, including a summary of findings related to the maturation of processes thought to facilitate segregation of target from competing speech. Data from published and ongoing studies are discussed, with a focus on synthesizing results from studies that address age-related changes in the ability to perceive speech in the presence of a small number of competing talkers. This review provides a summary of the current state of knowledge that is valuable for researchers and clinicians. It highlights the importance of considering listener factors, such as age and hearing status, as well as stimulus factors, such as masker type, when interpreting masked speech recognition data. http://cred.pubs.asha.org/article.aspx?articleid=2601620.