Signal Decomposition Technique Research Articles

Metaheuristic optimized time–frequency features for enhancing Alzheimer’s disease identification

Background:Alzheimer’s disease (AD) is a chronic disorder characterized by progressive cognitive dysfunctions and memory loss. Electroencephalography (EEG) is a non-invasive tool to detect AD using machine learning models and signal-derived features. Feature selection enhances AD detection performance, reduces training and testing time, and builds simpler models. Metaheuristic algorithms (MHAs) have successfully selected optimum features in several classification tasks and can be valuable in enhancing AD detection. Method:In this work, an AD detection model is proposed based on a systematic investigation of adaptive signal decomposition techniques and MHAs. The four techniques comprising empirical mode decomposition, variational mode decomposition, discrete wavelet transform, and Low-Complexity Orthogonal Wavelet Filter Banks (LCOWFBs) decompose EEG signals into subbands. Thirty-four features based on signal, temporal, spectral, and entropy analysis are extracted for each subband. The salient features are selected using seven different MHAs: particle swarm algorithm, salp swarm algorithm, reptile search (RSA), Grey Wolf, dwarf mongoose optimization, snake optimizer, and Fick’s law algorithm (FLA). Results:The two publicly available AD EEG datasets validated the performance of the proposed framework. LCOWFBs achieved the highest binary classification accuracy of 99.72% using 15 salient features selected by RSA and multi-class accuracy of 88.92% with seven salient features selected by FLA. The present method investigates brain regions that are affected by AD. Conclusion:The results indicate that the combination of LCOWFBs and RSA achieved maximum AD detection performance for both datasets with shorter computational time than earlier reported methods in the literature. The present model can be extended to identify chronic disorders, including hypertension, sleep disorders, and Parkinson’s disease.

Advanced series decomposition with a gated recurrent unit and graph convolutional neural network for non-stationary data patterns

In this study, we present the EEG-GCN, a novel hybrid model for the prediction of time series data, adept at addressing the inherent challenges posed by the data's complex, non-linear, and periodic nature, as well as the noise that frequently accompanies it. This model synergizes signal decomposition techniques with a graph convolutional neural network (GCN) for enhanced analytical precision. The EEG-GCN approaches time series data as a one-dimensional temporal signal, applying a dual-layered signal decomposition using both Ensemble Empirical Mode Decomposition (EEMD) and GRU. This two-pronged decomposition process effectively eliminates noise interference and distills the complex signal into more tractable sub-signals. These sub-signals facilitate a more straightforward feature analysis and learning process. To capitalize on the decomposed data, a graph convolutional neural network (GCN) is employed to discern the intricate feature interplay within the sub-signals and to map the interdependencies among the data points. The predictive model then synthesizes the weighted outputs of the GCN to yield the final forecast. A key component of our approach is the integration of a Gated Recurrent Unit (GRU) with EEMD within the GCN framework, referred to as EEMD-GRU-GCN. This combination leverages the strengths of GRU in capturing temporal dependencies and the EEMD's capability in handling non-stationary data, thereby enriching the feature set available for the GCN and enhancing the overall predictive accuracy and stability of the model. Empirical evaluations demonstrate that the EEG-GCN model achieves superior performance metrics. Compared to the baseline GCN model, EEG-GCN shows an average R2 improvement of 60% to 90%, outperforming the other methods. These results substantiate the advanced predictive capability of our proposed model, underscoring its potential for robust and accurate time series forecasting.

Time series forecasting model for non-stationary series pattern extraction using deep learning and GARCH modeling

This paper presents a novel approach to time series forecasting, an area of significant importance across diverse fields such as finance, meteorology, and industrial production. Time series data, characterized by its complexity involving trends, cyclicality, and random fluctuations, necessitates sophisticated methods for accurate forecasting. Traditional forecasting methods, while valuable, often struggle with the non-linear and non-stationary nature of time series data. To address this challenge, we propose an innovative model that combines signal decomposition and deep learning techniques. Our model employs Generalized Autoregressive Conditional Heteroskedasticity (GARCH) for learning the volatility in time series changes, followed by Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for data decomposition, significantly simplifying data complexity. We then apply Graph Convolutional Networks (GCN) to effectively learn the features of the decomposed data. The integration of these advanced techniques enables our model to fully capture and analyze the intricate features of time series data at various interval lengths. We have evaluated our model on multiple typical time-series datasets, demonstrating its enhanced predictive accuracy and stability compared to traditional methods. This research not only contributes to the field of time series forecasting but also opens avenues for the application of hybrid models in big data analysis, particularly in understanding and predicting the evolution of complex systems.

Multiple decomposition‐aided long short‐term memory network for enhanced short‐term wind power forecasting

AbstractWith the increasing penetration of grid‐scale wind energy systems, accurate wind power forecasting is critical to optimizing their integration into the power system, ensuring operational reliability, and enabling efficient system asset utilization. Addressing this challenge, this study proposes a novel forecasting model that combines the long‐short‐term memory (LSTM) neural network with two signal decomposition techniques. The EMD technique effectively extracts stable, stationary, and regular patterns from the original wind power signal, while the VMD technique tackles the most challenging high‐frequency component. A deep learning‐based forecasting model, i.e. the LSTM neural network, is used to take advantage of its ability to learn from longer sequences of data and its robustness to noise and outliers. The developed model is evaluated against LSTM models employing various decomposition methods using real wind power data from three distinct offshore wind farms. It is shown that the two‐stage decomposition significantly enhances forecasting accuracy, with the proposed model achieving values up to 9.5% higher than those obtained using standard LSTM models.

A water quality prediction model based on signal decomposition and ensemble deep learning techniques.

Accurate water quality predictions are critical for water resource protection, and dissolved oxygen (DO) reflects overall river water quality and ecosystem health. This study proposes a hybrid model based on the fusion of signal decomposition and deep learning for predicting river water quality. Initially, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is employed to split the internal series of DO into numerous internal mode functions (IMFs). Subsequently, we employed multi-scale fuzzy entropy (MFE) to compute the entropy values for each IMF component. Time-varying filtered empirical mode decomposition (TVFEMD) is used to further extract features in high-frequency subsequences after linearly aggregating the high-frequency sequences. Finally, support vector machine (SVM) and long short-term memory (LSTM) neural networks are used to predict low- and high-frequency subsequences. Moreover, by comparing it with single models, models based on 'single layer decomposition-prediction-ensemble' and combination models using different methods, the feasibility of the proposed model in predicting water quality data for the Xinlian section of Fuhe River and the Chucha section of Ganjiang River was verified. As a result, the combined prediction approach developed in this work has improved generalizability and prediction accuracy, and it may be used to forecast water quality in complicated waters.

Deep temporal networks for EEG-based motor imagery recognition.

The electroencephalogram (EEG) based motor imagery (MI) signal classification, also known as motion recognition, is a highly popular area of research due to its applications in robotics, gaming, and medical fields. However, the problem is ill-posed as these signals are non-stationary and noisy. Recently, a lot of efforts have been made to improve MI signal classification using a combination of signal decomposition and machine learning techniques but they fail to perform adequately on large multi-class datasets. Previously, researchers have implemented long short-term memory (LSTM), which is capable of learning the time-series information, on the MI-EEG dataset for motion recognition. However, it can not model very long-term dependencies present in the motion recognition data. With the advent of transformer networks in natural language processing (NLP), the long-term dependency issue has been widely addressed. Motivated by the success of transformer algorithms, in this article, we propose a transformer-based deep learning neural network architecture that performs motion recognition on the raw BCI competition III IVa and IV 2a datasets. The validation results show that the proposed method achieves superior performance than the existing state-of-the-art methods. The proposed method produces classification accuracy of 99.7% and 84% on the binary class and the multi-class datasets, respectively. Further, the performance of the proposed transformer-based model is also compared with LSTM.

Curvempirical Transform for Multimodal Fusion of Brain Images

Aims: Medical imaging requires special operating procedures and can cause mis-images that occur when someone is getting imaged, which can lead to inaccurate results Background: Adaptive illustration of the signal is imperative in signal processing. Empirical Wavelet Transform (EWT) is a new-fangled adaptive signal decomposition technique. Objective: Brain image fusion understands a dynamic job in medical imaging applications by assisting radiologists in detecting the variation in CT and MR images. Methods: This paper presents a fusion of filter banks of CT-MR image modalities of the Brain using the Empirical Curvelet Transform and Hybrid technique. In the hybrid technique filter banks of CT curvelet-MR little wood and CT little wood -MR curvelet were fused. The images were preprocessed using the Top Hat transform technique. The evaluation was performed based on the performance evaluation parameter. PSNR and SSIM are considered performance evaluation parameters Results: It has been observed that the results of fused filter banks using the curvelet technique show remarkable results in terms of PSNR and SSIM. The fused results show 29.10 dB PSNR and 0.819 SSIM. Conclusion: It has been observed that the fusion using only curvelet results in a 47.25% improvement in comparison with CT curvelet-MR little wood and a 42.68% improvement in comparison with CT little wood -MR curvelet.

An attention mechanism based deep nonlinear ensemble paradigm of strengthened feature extraction method for wind power prediction

The inherent uncertainty of wind power always hampers difficulties in the development of wind energy and the smooth operation of power systems. Therefore, reliable ultra-short-term wind power prediction is crucial for the development of wind energy. In this research, a two-stage nonlinear ensemble paradigm based on double-layer decomposition technology, feature reconstruction, intelligent optimization algorithm, and deep learning is suggested to increase the prediction accuracy of ultra-short-term wind power. First, using two different signal decomposition techniques for processing can further filter out noise in the original signal and fully capture different features within it. Second, the multiple components obtained through double decomposition are reconstructed using sample entropy theory and reassembled into several feature subsequences with similar complexity to simplify the input variables of the prediction model. Finally, based on the idea of a two-stage prediction strategy, the cuckoo search algorithm and the attention mechanism optimized long- and short-term memory model are applied to the prediction of feature subsequences and nonlinear integration, respectively, to obtain the final prediction results. Two sets of data from wind farms in Liaoning Province, China are used for simulation experiments. The final empirical findings indicate that, in comparison to other models, the suggested wind power prediction model has a greater prediction accuracy.

Boosting flood routing prediction performance through a hybrid approach using empirical mode decomposition and neural networks: a case study of the Mera River in Ankara

Abstract Flood routing is vital in helping to reduce the impact of floods on people and communities by allowing timely and appropriate responses. In this study, the empirical mode decomposition (EMD) signal decomposition technique is combined with cascade forward backpropagation neural network (CFBNN) and feed-forward backpropagation neural network (FFBNN) machine learning (ML) techniques to model 2014 floods in Ankara, Mera River. The data are split in order to avoid the underfitting and overfitting problems of the algorithm. While establishing the algorithm, 70% of the data were divided into training, 15% testing and 15% validation. Graphical indicators and statistical parameters were used for the analysis of model performance. As a result, the EMD signal decomposition technique has been found to improve the performance of ML models. In addition, the EMD-FFBNN hybrid model showed the most accurate estimation results in the flood routing calculation. The study's outputs can assist in designing flood control structures such as levees and dams to help reduce flood risk.

Optimized hybrid ensemble learning approaches applied to very short-term load forecasting

The significance of accurate short-term load forecasting (STLF) for modern power systems’ efficient and secure operation is paramount. This task is intricate due to cyclicity, non-stationarity, seasonality, and nonlinear power consumption time series data characteristics. The rise of data accessibility in the power industry has paved the way for machine learning (ML) models, which show the potential to enhance STLF accuracy. This paper presents a novel hybrid ML model combining Gradient Boosting Regressor (GBR), Extreme Gradient Boosting (XGBoost), k-Nearest Neighbors (kNN), and Support Vector Regression (SVR), examining both standalone and integrated, coupled with signal decomposition techniques like STL, EMD, EEMD, CEEMDAN, and EWT. Through Automated Machine Learning (AutoML), these models are integrated and their hyperparameters optimized, predicting each load signal component using data from two sources: The National Operator of Electric System (ONS) and the Independent System Operators New England (ISO-NE), boosting prediction capacity. For the 2019 ONS dataset, combining EWT and XGBoost yielded the best results for very short-term load forecasting (VSTLF) with an RMSE of 1,931.8 MW, MAE of 1,564.9 MW, and MAPE of 2.54%. These findings highlight the necessity for diverse approaches to each VSTLF problem, emphasizing the adaptability and strength of ML models combined with signal decomposition techniques.

A novel deformation monitoring model for high arch dams using impulse response-based equivalent temperature and machine learning-aided separate modeling

The anatomy of service status through health monitoring models is essential to long-term structural safety. Since deformation is the intuitive representation of the dam's operating condition, it is crucial to investigate the deformation monitoring model with high accuracy and strong interpretability for the safety management of high arch dams. A novel deformation monitoring model is proposed by incorporating impulse response-based equivalent temperature (ET) and machine learning-aided separate modeling technique (SMT). This methodology has three main sources of novelty. First, the impulse response-based equivalent temperature and corresponding temperature component are derived from heat transfer and temperature convolution theories. Then, the model parameters are identified by an improved firefly algorithm. Second, the components in the monitored displacement are progressively stripped out by clustering analysis, separation under equal water level conditions, and a robust signal decomposition technique. Third, the separated deformation components and environmental factors are modeled by the multi-output deep extreme learning machine (DELM) with autoencoder, and the monitoring model ET-SMT-DELM is thus established. The world's highest arch dam is selected to illustrate the proposed model, and the prediction accuracy, early warning performance, component shares, and impulse response mechanism are comprehensively investigated by comparing with several typical baseline models. The results show that the overfitting of the proposed model is reduced, and the prediction and early warning performance is significantly improved. The resulting temperature impulse response function and component shares imply that the interpretability of the model is also enhanced.

Multi-Objective Sparsity Maximum Mode De-Composition: A New Method for Rotating Machine Fault Diagnosis on High-Speed Train Axle Box

Signal decomposition techniques generally have two limitations. First, for adaptive correlated kurtogram (ACK) and variational mode decomposition (VMD), etc., the definition of the decomposition target has no direct relationship with the features of fault impacts, i.e., the decomposition is not guided by the health indices used to characterize bearing faults, resulting in fault impacts cannot be completely separated from vibration data. Second, it is hard to accurately determine the decomposition stop parameter (the number of final output modes), resulting in frequent over- or under-decomposition. All these are not conducive to bearing fault detection in engineering practice. In this paper, a new decomposition method, multi-objective sparse maximum mode decomposition (MOSMMD), is proposed. Firstly, by analyzing the Fourier spectrum of vibration data, MOSMMD constructs multiple finite impulse response filters covering different frequency bands to obtain initial candidate modes. Then, aiming at maximizing <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L2</i> / <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L1</i> norm of envelope spectrum (ES), the upper and lower cutoff frequencies of each filter are adjusted adaptively through iterative optimization, and the frequency bands covered by each mode are adjusted accordingly. Finally, by using that fault feature harmonics have periodic natures in ES, the redundant modes are eliminated and only the modes with obvious fault features are output. Compared with VMD and ACK, the decomposition target of MO-SMMD is directly related to the fault features, i.e., cyclostationarity of fault impacts, which makes the extraction of fault impacts more accurate. Moreover, the number of final output modes of MOSMMD is determined adaptively and does not need to be set in advance as an input parameter. Also, MOSMMD can decompose fault modes from vibration data without prior fault information, even when processing concurrent fault signals. Simulation and experiments demonstrate the superiority of the proposed MOSMMD.

Atrial fibrillation detection with signal decomposition and dilated residual neural network

Objective. Detecting atrial fibrillation (AF) using electrocardiogram (ECG) recordings from wearable devices has been challenging due to factors such as low signal-to-noise ratio and the use of only one lead. The use of deep learning has become a popular approach to tackle this task. However, it has been observed that current methods based on deep neural networks tend to favor raw signals as input, disregarding the valuable clinical experience in ECG diagnosis. Approach. In this study, we proposed a novel feature extraction method that generates a pseudo QRS complex signal and a pseudo T, P wave signal for each raw ECG signal using a temporal mask built upon R peak detection. Then a novel dilated residual neural network was trained on the decomposed signal. Main results. We evaluated the performance of our method on the dataset of PhysioNet/CinC 2017 Challenge, achieving an average score of 0.843. The method was further tested on MIT-BIH Atrial Fibrillation Database, and an average score of 0.984 was obtained. Significance. Our proposed ECG signal decomposition technique introduces simple and reliable domain knowledge into deep neural networks, and the dilated residual network provides large and flexible receptive fields, thereby enhancing the performance in the detection of AF. Our method can be extended to many other tasks involving ECG signals.

Automatic modal identification of bridges based on free vibrations and advanced signal decomposition techniques

Free vibration tests are attractive because they can be performed by means of a network consisting of few sensors temporarily installed on the structure in such a way to limit duration and cost of the experimental campaign. Additionally, free vibration tests are usually performed by introducing an initial perturbation that can induce a structural response significantly higher than the ambient excitation. This, in turn, allows to reduce the noise-to-signal ratio in the final measurements and/or to consider less stringent requirements about the technical specifications of the sensors. Since free vibration tests can provide accurate estimates of the modal parameters while being rather cheap and easy to implement, they have been performed in many applications, such as the experimental dynamic characterization of base-isolated buildings, masonry towers, ancient tie-rods, and bridges. An efficient and automatic computational framework is thus presented for the modal identification of bridges based on their free vibrations. The novel procedure proposed in the current work combines advanced signal decomposition techniques and a robust approach for damping identification. Two advanced signal decomposition techniques are considered, namely the Variational Mode Decomposition and the Empirical Fourier Decomposition. Experimental applications are then illustrated for roadway and railway bridges.

A Review of Methods and Applications for a Heart Rate Variability Analysis

Heart rate variability (HRV) has emerged as an essential non-invasive tool for understanding cardiac autonomic function over the last few decades. This can be attributed to the direct connection between the heart’s rhythm and the activity of the sympathetic and parasympathetic nervous systems. The cost-effectiveness and ease with which one may obtain HRV data also make it an exciting and potential clinical tool for evaluating and identifying various health impairments. This article comprehensively describes a range of signal decomposition techniques and time-series modeling methods recently used in HRV analyses apart from the conventional HRV generation and feature extraction methods. Various weight-based feature selection approaches and dimensionality reduction techniques are summarized to assess the relevance of each HRV feature vector. The popular machine learning-based HRV feature classification techniques are also described. Some notable clinical applications of HRV analyses, like the detection of diabetes, sleep apnea, myocardial infarction, cardiac arrhythmia, hypertension, renal failure, psychiatric disorders, ANS Activity of Patients Undergoing Weaning from Mechanical Ventilation, and monitoring of fetal distress and neonatal critical care, are discussed. The latest research on the effect of external stimuli (like consuming alcohol) on autonomic nervous system (ANS) activity using HRV analyses is also summarized. The HRV analysis approaches summarized in our article can help future researchers to dive deep into their potential diagnostic applications.

An improved variational mode decomposition method based on spectrum reconstruction and segmentation and its application in rolling bearing fault diagnosis

Variational mode decomposition (VMD) is an adaptable signal decomposition approach. VMD has excellent noise abatement performance, which is frequently used to identify rolling bearing failures. The primary problem with VMD is the incapacity to identify the number of modes in an adaptive way, which can seriously affect the decomposition effect. To overcome this problem, an improved VMD method based on spectrum reconstruction and segmentation (SRAS-VMD) is presented in this paper. First, the spectrum is simplified by spectrum reconstruction and its energy spectrum is extracted for spectrum pre-segmentation. Then, the boundary fusion is performed according to Gini index of the squared envelope. The initial center frequency and the number of modes are identified from the fusion results. Finally, the optimal mode is selected using the periodic modulation intensity. The envelope spectrum of the optimal mode is observed, and fault diagnosis is performed. Simulation analyses and experimental signals of bearings studies demonstrate the feasibility of this scheme in diagnosing both single point and composite fault of bearings. The advantageousness of this method is shown by contrast with currently available adaptive signal decomposition techniques.

Application of novel artificial bee colony optimized ANN and data preprocessing techniques for monthly streamflow estimation.

Streamflow estimation is important in hydrology, especially in drought and flood-prone areas. Accurate estimation of streamflow values is crucial for the sustainable management of water resources, the development of early warning systems for disasters, and for various applications such as irrigation, hydropower production, dam sizing, and siltation management. This study developed the ANN algorithm by optimizing with an artificial bee colony (ABC). Then, the ABC-ANN hybrid model, which was established, was combined with different signal decomposition techniques to evaluate its performance in streamflow estimation in the East Black Sea Region, Türkiye. For this purpose, the lagged streamflow values were divided into subcomponents using the local mean decomposition (LMD) with the empirical envelope and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) signal decomposition techniques presented to the ABC-ANN algorithm. Thus, the success of the novel hybrid LMD-ABC-ANN and CEEMDAN-ABC-ANN approaches in streamflow prediction was evaluated. The outputs are reliable strategies and resources for water resource planners and policymakers.

Efficacy of Vehicle Scanning Methods in Estimating the Mode Shapes of Bridges Seated on Elastic Supports.

This study systematically assesses the efficacy of the vehicle scanning methods (VSM) in accurately estimating the mode shapes of bridges seated on elastic supports. Three state-of-the-art VSM methods are employed to obtain the mode shapes of bridges using the vehicle data during its travel. Two of the evaluated methods use a signal decomposition technique to extract the modal components of the bridge from the contact point of the response while the third one relies on the segmentation of the measured signals along the bridge deck and applying an operational modal analysis tool to each segmented signal to estimate the mode shapes. Numerical analyses are conducted on one single- and one two-span bridge, considering smooth and rough road profiles, different vehicle speeds, and presence of lead vehicle. The accuracy of the numerical models used in developing and assessing vehicle scanning models is tested, and the results of the study demonstrate the method using a half-car vehicle model and signal decomposition technique shows robustness against increasing vehicle speeds and road roughness while the method applying the segmentation of the measured signals provides relatively accurate mode shape estimates at the bridge edges at low speed, although the three methods have their limitations. It is also observed that simplified bridge and vehicle models can hide potential challenges that arise from the complexity of actual vehicle and bridge systems. Considering that a significant number of bridges worldwide are built on elastic supports, the practical success of vehicle scanning methods depends on their ability to handle elastic boundary conditions with reliability. Therefore, the article provides valuable insights into the capabilities and limitations of the current vehicle scanning methods, paving the way for further advancements and refinements in these techniques.

Synergy biochemical and physiological traits for rice heavy metal stress detection from Sentinel-2 images using signal–decomposition technique

The increasing pollution of soil with heavy metals has greatly damaged the ecological environment and endangered human health. Heavy metal stress affects not only biochemical constituents (i.e., chlorophyll) but also physiological functions (i.e., photosynthesis). We aimed to detect heavy metal stress in rice by integrating these two aspects. We selected Zhuzhou, Hunan Province as the study area and collected Sentinel-2 images from 2019 to 2021 and situ-measured data. First, two indicators were computed from Sentinel-2 images characterizing biochemical components and physiological functions, respectively. Second, the original time series were decomposed using the complete ensemble empirical mode decomposition with adaptive noise. Finally, a time-spectrum feature space (TSFS) model based on signal decomposition in critical periods sensitive to heavy metal stress was developed to determine the level of heavy metal stress. It showed that, regardless of individuals years, the overall accuracy and kappa coefficients of the TSFS model were >85 % and 0.8, and compared with a single indicator, combining two indicators with the signal-decomposition method significantly improved the accuracy of stress detection. This demonstrates that the integration of biochemical components with physiological functions and signal decomposition techniques can provide an important reference for crop heavy metal stress monitoring.

Recognizing Command Words using Deep Recurrent Neural Network for Both Acoustic and Throat Speech

The importance of speech command recognition in a human-machine interaction system is increased in recent years. In this study, we propose a deep neural network-based system for acoustic and throat command speech recognition. We apply a preprocessed pipeline to create the input of the deep learning model. Firstly, speech commands are decomposed into components using well-known signal decomposition techniques. The Mel-frequency cepstral coefficients (MFCC) feature extraction method is applied to each component of the speech commands to obtain the feature inputs for the recognition system. At this stage, we apply and compare performance using different speech decomposition techniques such as wavelet packet decomposition (WPD), continuous wavelet transform (CWT), and empirical mode decomposition (EMD) in order to find out the best technique for our model. We observe that WPD shows the best performance in terms of classification accuracy. This paper investigates long short-term memory (LSTM)-based recurrent neural network (RNN), which is trained using the extracted MFCC features. The proposed neural network is trained and tested using acoustic speech commands. Moreover, we also train and test the proposed model using a throat mic. speech commands as well. Lastly, the transfer learning technique is employed to increase the test accuracy for throat speech recognition. The weights of the model train with the acoustic signal are used to initialize the model used for throat speech recognition. Overall, we have found significant classification accuracy for both acoustic and throat command speech. We obtain LSTM is much better than the GMM-HMM model, convolutional neural networks such as CNN-tpool2 and residual networks such as res15 and res26 with an accuracy score of over 97% on Google’s Speech Commands dataset and we achieve 95.35% accuracy on our throat speech data set using the transfer learning technique.