Digit Speech Recognition Research Articles

Crossmixed convolutional neural network for digital speech recognition.

Digital speech recognition is a challenging problem that requires the ability to learn complex signal characteristics such as frequency, pitch, intensity, timbre, and melody, which traditional methods often face issues in recognizing. This article introduces three solutions based on convolutional neural networks (CNN) to solve the problem: 1D-CNN is designed to learn directly from digital data; 2DS-CNN and 2DM-CNN have a more complex architecture, transferring raw waveform into transformed images using Fourier transform to learn essential features. Experimental results on four large data sets, containing 30,000 samples for each, show that the three proposed models achieve superior performance compared to well-known models such as GoogLeNet and AlexNet, with the best accuracy of 95.87%, 99.65%, and 99.76%, respectively. With 5-10% higher performance than other models, the proposed solution has demonstrated the ability to effectively learn features, improve recognition accuracy and speed, and open up the potential for broad applications in virtual assistants, medical recording, and voice commands.

BCM‐Inspired Synapses Constructed with Barrier‐Modulated Coupling Junctions for Enhancing Speech Recognition

AbstractBio‐inspired synaptic devices have garnered considerable interest in neuromorphic computing. The Bienenstock‐Cooper‐Munro (BCM) learning rule stands out as one of the most accurate synaptic models, featuring non‐monotonic behavior and threshold sliding effect, crucial for stable learning processes. The direct device strategy for completely mimicking the BCM rule is a tough issue since the current devices lack two competitive working modes within one device. In this work, a dual‐junction synaptic device with opposite built‐in electric fields using a W/WO2/WO3‐x/Au structure is demonstrated. The devices directly mimic two fundamental features of the BCM rule via a delicately‐designed bandgap engineering strategy. Furthermore, the working mechanisms are investigated and the promising potential of dual‐junction synaptic devices is demonstrated for enhancing speech recognition through Convolutional Neural Network (CNN)‐based digital speech recognition with a remarkable accuracy of 98% through a synaptic array. Even for speech recognition with 13% Gaussian noise, the accuracy remained at 83%. These findings provide a promising strategy for developing BCM‐based synaptic devices for neuromorphic computing applications.

Deep Learning-Based Audio-Visual Speech Recognition for Bosnian Digits

This study presents a deep learning-based solution for audio-visual speech recognition of Bosnian digits. The task posed a challenge due to the lack of an appropriate Bosnian language dataset, and this study outlines the approach to building a new dataset. The proposed solution includes two components: visual speech recognition, which involves lip reading, and audio speech recognition. For visual speech recognition, a combined CNN-RNN architecture was utilised, consisting of two CNN variants namely Google Net and ResNet-50. These architectures were compared based on their performance, with ResNet-50 achieving 72% accuracy and Google Net achieving 63% accuracy. The RNN component used LSTM. For audio speech recognition, FFT is applied to obtain spectrograms from the input speech signal, which are then classified using a CNN architecture. This component achieved an accuracy of 100%. The dataset was split into three parts namely for training, validation, and testing purposes such that 80%, 10% and 10% of data is allocated to each part, respectively. Furthermore, the predictions from the visual and audio models were combined that yielded 100% accuracy based on the developed dataset. The findings from this study demonstrate that deep learning-based methods show promising results for audio-visual speech recognition of Bosnian digits, despite the challenge of limited Bosnian language datasets.

Experimental Analysis on Performance of Speech Utterance recognition using AI Models

In Automated speech recognition of the system performance is crucial and important to satisfy multiple requirements of HMI and, more recently, even in IoT-related applications as well. Concurrently, there has been an increase in demand for detecting strong critical features derived from speech utterances. This paper presents a performance of the developed machine learning algorithms with respect to audio digit speech recognition and classification. The prepared dataset contains a free range of words (from 1 to 10) from speakers of different age groups. The Audacity software used for preprocessing the audio files that includes removal of noise included in the signal and trimming the silence on either side of the word utterance. audio signal sampled at fs = 48kHz.We have developed four AI Models to recognise the word utterances. Audio signals are processed separately and derived two unique feature sets that includes statistical features set and singular values by performing SVD related to word utterances. The cepstral values for each utterance are obtained from state-of-the-art MFCC. Variance-covariance matrix is calculated from the generated MFCC matrix. The diagonal values which form the variance are recorded and denoted as feature set-1 for the word utterance and inputted to the machine learning algorithms. Performance matrices of the developed models are recorded. To keep the computational bottleneck associated with the use of feature sets to minimum, dimensionality reduction is carried out by applying singular value decomposition to the extracted MFCC matrix. The derived set of singular values ​​considered as feature set-2 is used to train and test the developed AI models with a ratio of 70:30. We presented and discussed the performance and results produced by MLP, KNN, SVM, Random Forest algorithms. In comparison, MLP and Random Forest were found to show excellent performance on both feature sets with 100% training accuracy and 99% test accuracy.

A novel learning approach in deep spiking neural networks with multi-objective optimization algorithms for automatic digit speech recognition

A novel learning approach in deep spiking neural networks with multi-objective optimization algorithms for automatic digit speech recognition

Automatic speech recognition of Gujarati digits using wavelet coefficients in machine learning algorithms

Automatic speech recognition of Gujarati digits using wavelet coefficients in machine learning algorithms

Automatic speech recognition of Gujarati digits using wavelet coefficients in machine learning algorithms

Automatic speech recognition of Gujarati digits using wavelet coefficients in machine learning algorithms

Classification SARS-CoV-2 Disease based on CT-Scan Image Using Convolutional Neural Network

Purpose: Convolutional Neural Network (CNN) is one of the most popular and widely used deep learning algorithms. These algorithms are commonly used in various applications, including image processing in medical and digital forensics, speech recognition, and other academic disciplines. SARS-CoV-2 (COVID-19) is a disease that first appeared in Wuhan, China, and has symptoms similar to pneumonia. This study aims to classify the covid-19 virus by proposing a deep learning model to prevent infection rates.Methods: The dataset used in this study is a public dataset originating from a hospital in Sao Paulo, Brazil. The data images consisted of 1252 infected with covid and 1230 data classified as non-covid but have other lung diseases. The classification method proposed in this research is a CNN model based on Resnet 50.Result: The experimental results show that the proposed Resnet 50-based convolutional neural network model works well in classifying SARS-CoV-2 disease using CT-Scan images. Our proposed model obtains 95% accuracy, precision, recall, and f1 values on the Epoch 500.Novelty: In this experiment, we utilized the Resnet50-based CNN model to classify the SARS-CoV-2 (COVID-19) disease using CT-Scan images and got good performance.

Use Brain-Like Audio Features to Improve Speech Recognition Performance

Speech recognition plays an important role in the field of human-computer interaction through the use of acoustic sensors, but speech recognition is technically difficult, has complex overall logic, relies heavily on neural network algorithms, and has extremely high technical requirements. In speech recognition, feature extraction is the first step in speech recognition for recovering and extracting speech features. Existing methods, such as Meier spectral coefficients (MFCC) and spectrograms, lose a large amount of acoustic information and lack biological interpretability. Then, for example, existing speech self-supervised representation learning methods based on contrast prediction need to construct a large number of negative samples during training, and their learning effects depend on large batches of training, which requires a large amount of computational resources for the problem. Therefore, in this paper, we propose a new feature extraction method, called SHH (spike-H), that resembles the human brain and achieves higher speech recognition rates than previous methods. The features extracted using the proposed model are subsequently fed into the classification model. We propose a novel parallel CRNN model with an attention mechanism that considers both temporal and spatial features. Experimental results show that the proposed CRNN achieves an accuracy of 94.8% on the Aurora dataset. In addition, audio similarity experiments show that SHH can better distinguish audio features. In addition, the ablation experiments show that SHH is applicable to digital speech recognition.

Digital Music Feature Recognition Based on Wireless Sensing Technology

With the rapid development of information technology, digital music is subsequently increasing in large quantities, and how a good integration of vocal input and recognition technology can be transformed into digital music can greatly improve the efficiency of music production while ensuring the quality and effect of music. This paper focuses on the implementation and application of human voice input and recognition technology in digital music creation, enabling users to generate digital music forms by simply humming a melodic fragment of a piece of music into a microphone. The paper begins with an introduction to digital music and speech recognition technology and goes on to describe the respective characteristics of various audio formats, which are selected as data sources for digital music creation based on the advantages of the files in terms of retrieval. Following that, the method of extracting musical information from music is described, and the main melody is successfully extracted from the multitrack file to extract the corresponding musical performance information. The feature extraction of humming input melody is further described in detail. The traditional speech recognition method of using short-time energy and short-time overzero rate features for speech endpoint detection is analyzed. Combining the characteristics of humming music, the method of cutting notes by two-stage cutting mode, i.e., combining energy saliency index, overzero rate, and pitch change, is adopted to cut notes, which leads to a substantial improvement in performance. The algorithm uses the melody extraction algorithm to obtain the melody line, merges the short-time segments of the melody line to reduce the error rate of emotion recognition, uses the melody line to segment the music signal to generate segmented segments, then abstracts the features of the segmented segments through a CNN-based structural model, and inputs the output of the model to the regressor in cascade with the melody contour features of the corresponding segmented segments to finally obtain the emotion V / A value of the segmented segments.

Speech recognition for medical documentation: an analysis of time, cost efficiency and acceptance in a clinical setting

Background/Aims Medical documentation is an important and unavoidable part of a health professional's working day. However, the time required for medical documentation is often viewed negatively, particularly by clinicians with heavy workloads. Digital speech recognition has become more prevalent and is being used to optimise working time. This study evaluated the time and cost savings associated with speech recognition technology, and its potential for improving healthcare processes. Methods Clinicians were directly observed while completing medical documentation. A total of 313 samples were collected, of which 163 used speech recognition and 150 used typing methods. The time taken to complete the medical form, the error rate and error correction time were recorded. A survey was also completed by 31 clinicians to gauge their level of acceptance of speech recognition software for medical documentation. Two-sample t-tests and Mann–Whitney U tests were performed to determine statistical trends and significance. Results On average, medical documentation using speech recognition software took just 5.11 minutes to complete the form, compared to 8.9 minutes typing, representing significant time savings. The error rate was also found to be lower for speech recognition software. However, 55% of clinicians surveyed stated that they would prefer to type their notes rather than use speech recognition software and perceived the error rate of this software to be higher than typing. Conclusions The results showed that there are both temporal and financial advantages of speech recognition technology over text input for medical documentation. However, this technology had low levels of acceptance among staff, which could have implications for the uptake of this method.

Digits-in-noise test in Brazilian Portuguese: how demographic and socioeconomic variables influence normal-hearing subjects.

Verify how demographic and socioeconomic variables on the in-noise speech recognition threshold (SRT) from the digits-in-noise test (DIN) in Brazilian Portuguese influence normal-hearing subjects. Cross-sectional, prospective study. The convenience sample had 151 normal-hearing subjects between 12 and 79 years (mean=34.66) who underwent pure tone audiometry and digits-in-noise test with white noise using a sequence of three numbers in diotic stimulus (in-phase) on the same day. The DIN was performed using a Motorola Z3 Play smartphone with internet access and in-ear headphones. In-noise digit speech recognition threshold (SRT) was analyzed for gender, age, educational levels, and socioeconomic status. We used the non-parametric version of the Kruskal-Wallis and Mann-Whitney U tests to compare independent samples adopting a significance level of 5%. The mean SRT was -8.47 dBNA (SD -3.89) with a median of -9.6 dBNA. The SRT was proportionally inverse to educational levels and socioeconomic status and more negative (better) with lower age groups. Gender did not influence the DIN SRT. Age, educational levels, and socioeconomic status influenced the DIN threshold. These variables must be considered when analyzing DIN performance in Brazilian Portuguese in normal-hearing subjects.

Application of digital technology in the work of a pathologist: guidelines for learning how to use speech recognition systems

Natural language processing is one of the branches of computational linguistics. It is a branch of computer science that includes algorithmic processing of speech and natural language scripts. The algorithms facilitate the development of human-to-machine translation and automatic speech recognition systems (ASRS). ASRS use is twofold: accurately converting operators speech into a coherent and meaningful text and using natural language for interaction with a computer. Currently, these systems are widely used in medical practice, including anatomic pathology. Successful ASRS implementation pivots on creation of standardized templated descriptions for organic inclusion in the diagnosis dictation, likewise on physician training for using such systems in practice. In the past decade, there have been several attempts to standardize surgical pathology reports and create templates undertaken by physicians worldwide. After studying domestic and foreign literature, we created a list of the essential elements that must be included in the template for macro-and microscopic descriptions comprising the final diagnosis. These templates will help in decision-making and accurate diagnosis as they contain all the imperative elements in order of importance. This approach will significantly reduce the need for re-examination of both fixed macroscopic material and the preparation of additional histological sections. The templates built into ASRS reduce the time spent on documentation and significantly reduce the workload for pathologists. For the successful use of ASRS, we have developed an educational course, Digital Speech Recognition in an Anatomical Pathology Practice, for postgraduate education of both domestic and foreign doctors. A brief description of the course is presented in this article, and the course itself is available on the Internet.

Forward Digit Span and Word Familiarity Do Not Correlate With Differences in Speech Recognition in Individuals With Cochlear Implants After Accounting for Auditory Resolution.

Purpose In individuals with cochlear implants, speech recognition is not associated with tests of working memory that primarily reflect storage, such as forward digit span. In contrast, our previous work found that vocoded speech recognition in individuals with normal hearing was correlated with performance on a forward digit span task. A possible explanation for this difference across groups is that variability in auditory resolution across individuals with cochlear implants could conceal the true relationship between speech and memory tasks. Here, our goal was to determine if performance on forward digit span and speech recognition tasks are correlated in individuals with cochlear implants after controlling for individual differences in auditory resolution. Method We measured sentence recognition ability in 20 individuals with cochlear implants with Perceptually Robust English Sentence Test Open-set sentences. Spectral and temporal modulation detection tasks were used to assess individual differences in auditory resolution, auditory forward digit span was used to assess working memory storage, and self-reported word familiarity was used to assess vocabulary. Results Individual differences in speech recognition were predicted by spectral and temporal resolution. A correlation was found between forward digit span and speech recognition, but this correlation was not significant after controlling for spectral and temporal resolution. No relationship was found between word familiarity and speech recognition. Forward digit span performance was not associated with individual differences in auditory resolution. Conclusions Our findings support the idea that sentence recognition in individuals with cochlear implants is primarily limited by individual differences in working memory processing, not storage. Studies examining the relationship between speech and memory should control for individual differences in auditory resolution.

Metode Wavelet-MFCC dan Korelasi dalam Pengenalan Suara Digit

Voice is the sound emitted from living things. With the development of Automatic Speech Recognition (ASR) technology, voice can be used to make it easier for humans to do something. In the ASR extraction process the features have an important role in the recognition process. The feature extraction methods that are commonly applied to ASR are MFCC and Wavelet. Each of them has advantages and disadvantages. Therefore, this study will combine the wavelet feature extraction method and MFCC to maximize the existing advantages. The proposed method is called Wavelet-MFCC. Voice recognition method that does not use recommendations. Determination of system performance using the Word Recoginition Rate (WRR) method which is validated with the K-Fold Cross Validation with the number of folds is 5. The research dataset used is voice recording digits 0-9 in English. The results show that the digit speech recognition system that has been built gives the highest average value of 63% for digit 4 using wavelet daubechies DB3 and wavelet dyadic transform method. As for the comparison results of the wavelet decomposition method used, that the use of dyadic wavelet transformation is better than the wavelet package.

Arabic digits speech recognition and speaker identification in noisy environment using a hybrid model of VQ and GMM

This paper presents an automatic speaker identification and speech recognition for Arabic digits in noisy environment. In this work, the proposed system is able to identify the speaker after saving his voice in the database and adding noise. The mel frequency cepstral coefficients (MFCC) is the best approach used in building a program in the Matlab platform; also, the quantization is used for generating the codebooks. The Gaussian mixture modelling (GMM) algorithms are used to generate template, feature-matching purpose. In this paper, we have proposed a system based on MFCC-GMM and MFCC-VQ Approaches on the one hand and by using the Hybrid Approach MFCC-VQ-GMM on the other hand for speaker modeling. The White Gaussian noise is added to the clean speech at several signal-to-noise ratio (SNR) levels to test the system in a noisy environment. The proposed system gives good results in recognition rate.

Digit Speech Recognition u sing Hidden Markov Model Toolkit

Digit speech recognition refers to the task of identifying the English digit spoken in a particular utterance by an unknown speaker. The conventional methods used for the recognition of digits in speech are based on robust pattern recognition techniques which deal with the statistical parameters of speech. HMM, GMM and dynamic programming techniques are some of the methods. This paper presents recognition of digits using HTK Toolkit which is based on Hidden Markov Model using MFCC. The digit speech database of this work was collected in real time from both male and female speakers and the transcription of the total collected data was done using Wavesurfer.

A Multiple-Input Multiple-Output Reservoir Computing System Subject to Optoelectronic Feedbacks and Mutual Coupling.

In this paper, a multiple-input multiple-output reservoir computing (RC) system is proposed, which is composed of multiple nonlinear nodes (Mach–Zehnder modulators) and multiple mutual-coupling loops of optoelectronic delay lines. Each input signal is added into every mutual-coupling loop to implement the simultaneous recognition of multiple route signals, which results in the signal processing speed improving and the number of routes increasing. As an example, the four-route input and four-route output RC is simultaneously realized by numerical simulations. The results show that this type of RC system can successfully recognize the four-route optical packet headers with 3-bit, 8-bit, 16-bit, and 32-bit, and four-route independent digital speeches. When the white noise is added to the signals such that the signal-to-noise ratio (SNR) of the optical packet headers and the digital speeches are 35 dB and 20 dB respectively, the normalized root mean square errors (NRMSEs) of the signal recognition are all close to 0.1. The word error rates (WERs) of the optical packet header recognition are 0%. The WER of the digital speech recognition is 1.6%. The eight-route input and eight-route output RC is also numerically simulated. The recognition of the eight-route 3-bit optical packet headers is implemented. The parallel processing of multiple-route signals and the high recognition accuracy are implemented by this proposed system.

Added Value of Structured Reporting for Medical Practice and Management

Background: Structured reports are getting popular gradually. To increase the adaptation of the technology, we will briefly go over the benefits that structured reports can provide to almost all medical staff and the medical community in general. Objectives: Learning objectives include: 1. What are the benefits of SR for medical doctors? 2. What are the benefits of SR for patients? 3. How can SR boost high quality research? Outline: First, I will briefly go over some of the known benefits of SR, as follows: • Disease and domain-specific report templates can increase the clarity and quality of the report. • The use of common data elements ensures the consistent use of terminology across practices. • The use of checklists inherently in structured reports reduces diagnostic errors. • Less grammatical and nongrammatical errors may be introduced into SR even when digital speech recognition is used. • Preserving the completeness of report documentation improves insurance and other reimbursements. • It improves quality. • It may promote evidence-based medicine by integrating clinical decision support tools with radiology reports. However, the most important factor is to improve research. Each population based on genetic background and ethnicity may require different or specific medical protocols or practice for certain diseases. High quality medical research is needed to address the differences and to build the foundation for more appropriate medical procedures and knowledge generation. The importance of high impact and high-quality research in medicine and medical practice is felt in Iranian universities but irrespective a large amount of government investments on different aspects of medical fields is not clearly observable. The universities have abundant numbers of erudite and competent researchers but not enough tagged or labeled data are available for high impact publications. Medical doctors in Iran are mainly practitioners. Although research has gained momentum within the last few years, mainstream respected researchers in medicine do not put research in their first priority. Structured reporting, if performed properly, can provide the main feed for quality research since while medical practitioners perform their regular medical practice. Their diagnosis and observations can be used directly as input to data mining and machine learning algorithms and at the same time be used for population studies.

Efficient optoelectronic reservoir computing with three-route input based on optical delay lines.

In the reservoir computing (RC) system based on delay line structure, the dynamical reservoir has short-term memory capability and can map instantaneous time signals to a high-dimensional space, so that it can effectively classify and recognize complicated time signals. A new RC system is proposed in this paper, where three-way signals are simultaneously input into the dynamical reservoir, and simultaneously processed and classified. Therefore, the information processing speed of the RC system is promoted manyfold. The reservoir is composed of a semiconductor laser and two optical feedback loops, in which two input signals are injected into two optical feedback loops, and another input signal is added to the driving current of the semiconductor laser. The computation efficiency of the RC system is validated by two tasks of the optical packet header recognition and digital speech recognition. By optimizing the parameters, the multi-input RC system can simultaneously recognize three-way 3-bit, 8-bit, 16-bit, and 32-bit optical packet headers with the optimal word error rate (WER) of 0%, and recognize three-way digital speech signals with the optimal WER of 0.2%. It is proven that this multi-input system outperforms the single-inputting RC system based on delay line structure.