Mel-frequency Spectral Coefficients Research Articles

Assessing the Performance of a Speech Recognition System Embedded in Low-Cost Devices

The main purpose of this research is to investigate how an Amazigh speech recognition system can be integrated into a low-cost minicomputer, specifically the Raspberry Pi, in order to improve the system's automatic speech recognition capabilities. The study focuses on optimizing system parameters to achieve a balance between performance and limited system resources. To achieve this, the system employs a combination of Hidden Markov Models (HMMs), Gaussian Mixture Models (GMMs), and Mel Frequency Spectral Coefficients (MFCCs) with a speaker-independent approach. The system has been developed to recognize 20 Amazigh words, comprising of 10 commands and the first ten Amazigh digits. The results indicate that the recognition rate achieved on the Raspberry Pi system is 89.16% using 3 HMMs, 16 GMMs, and 39 MFCC coefficients. These findings demonstrate that it is feasible to create effective embedded Amazigh speech recognition systems using a low-cost minicomputer such as the Raspberry Pi. Furthermore, Amazigh linguistic analysis has been implemented to ensure the accuracy of the designed embedded speech system.

Heart sound classification based on improved mel-frequency spectral coefficients and deep residual learning.

Heart sound classification plays a critical role in the early diagnosis of cardiovascular diseases. Although there have been many advances in heart sound classification in the last few years, most of them are still based on conventional segmented features and shallow structure-based classifiers. Therefore, we propose a new heart sound classification method based on improved mel-frequency cepstrum coefficient features and deep residual learning. Firstly, the heart sound signal is preprocessed, and its improved features are computed. Then, these features are used as input features of the neural network. The pathological information in the heart sound signal is further extracted by the deep residual network. Finally, the heart sound signal is classified into different categories according to the features learned by the neural network. This paper presents comprehensive analyses of different network parameters and network connection strategies. The proposed method achieves an accuracy of 94.43% on the dataset in this paper.

COVID-19 assessment using HMM cough recognition system.

This paper is a part of our contributions to research on the ongoing COVID-19 pandemic around the world. This research aims to use Hidden Markov Model (HMM) based automatic speech recognition system to analyze the cough signal and determine whether the signal belongs to a sick or healthy speaker. We built a configurable model by using HMMs, Gaussian Mixture Models (GMMs), Mel frequency spectral coefficients (MFCCs) and a cough corpus collected from healthy and sick voluntary speakers. Our proposed method is able to classify dry cough with sensitivity from 85.86% to 91.57%, differentiate the dry cough, and cough COVID-19 symptom with specificity from 5 to 10%. The obtained results are very encouraging to enrich our corpus with more data and increase the performance of our diagnostic system.

A novel silent speech recognition approach based on parallel inception convolutional neural network and Mel frequency spectral coefficient.

Silent speech recognition breaks the limitations of automatic speech recognition when acoustic signals cannot be produced or captured clearly, but still has a long way to go before being ready for any real-life applications. To address this issue, we propose a novel silent speech recognition framework based on surface electromyography (sEMG) signals. In our approach, a new deep learning architecture Parallel Inception Convolutional Neural Network (PICNN) is proposed and implemented in our silent speech recognition system, with six inception modules processing six channels of sEMG data, separately and simultaneously. Meanwhile, Mel Frequency Spectral Coefficients (MFSCs) are employed to extract speech-related sEMG features for the first time. We further design and generate a 100-class dataset containing daily life assistance demands for the elderly and disabled individuals. The experimental results obtained from 28 subjects confirm that our silent speech recognition method outperforms state-of-the-art machine learning algorithms and deep learning architectures, achieving the best recognition accuracy of 90.76%. With sEMG data collected from four new subjects, efficient steps of subject-based transfer learning are conducted to further improve the cross-subject recognition ability of the proposed model. Promising results prove that our sEMG-based silent speech recognition system could have high recognition accuracy and steady performance in practical applications.

Analysis of constant-Q filterbank based representations for speech emotion recognition

This work analyzes the constant-Q filterbank-based time-frequency representations for speech emotion recognition (SER). Constant-Q filterbank provides non-linear spectro-temporal representation with higher frequency resolution at low frequencies. Our investigation reveals how the increased low-frequency resolution benefits SER. The time-domain comparative analysis between short-term mel-frequency spectral coefficients (MFSCs) and constant-Q filterbank-based features, namely constant-Q transform (CQT) and continuous wavelet transform (CWT), reveals that constant-Q representations provide higher time-invariance at low-frequencies. This provides increased robustness against emotion irrelevant temporal variations in pitch, especially for low-arousal emotions. The corresponding frequency-domain analysis over different emotion classes shows better resolution of pitch harmonics in constant-Q-based time-frequency representations than MFSC. These advantages of constant-Q representations are further consolidated by SER performance in the extensive evaluation of features over four publicly available databases with six advanced deep neural network architectures as the back-end classifiers. Our inferences in this study hint toward the suitability and potentiality of constant-Q features for SER.

Heart Sound Classification based on Residual Shrinkage Networks.

Heart sound classification is one of the non-invasive methods for early detection of the cardiovascular diseases (CVDs), the leading cause for deaths. In recent years, Computer Audition (CA) technology has become increasingly sophisticated, auxiliary diagnosis technology of heart disease based on CA has become a popular research area. This paper proposes a deep Convolutional Neural Network (CNN) model for heart sound classification. To improve the classification accuracy of heart sound, we design a classification algorithm combining classical Residual Network (ResNet) and Long Short-Term Memory (LSTM). The model performance is evaluated in the PhysioNet/CinC Challenges 2016 datasets using a 2D time-frequency feature. We extract the four features from different filter-bank coefficients, including Filterbank (Fbank), Mel-Frequency Spectral Coefficients (MFSCs), and Mel-Frequency Cepstral Coefficients (MFCCs). The experimental results show the MFSCs feature outperforms the other features in the proposed CNN model. The proposed model performs well on the test set, particularly the F1 score of 84.3 % - the accuracy of 84.4 %, the sensitivity of 84.3 %, and the specificity of 85.6 %. Compared with the classical ResNet model, an accuracy of 4.9 % improvement is observed in the proposed model.

Parallel-Inception CNN Approach for Facial sEMG based Silent Speech Recognition.

With the purpose of providing an external human-machine interaction platform for the elderly in need, a novel facial surface electromyography based silent speech recognition system was developed. In this study, we propose a deep learning architecture named Parallel-Inception Convolutional Neural Network (PICNN), and employ up-to-date feature extraction method log Mel frequency spectral coefficients (MFSC). To better meet the requirements of our target users, a 100-class dataset containing daily life-related demands was designed and generated for the comparative experiments. According to experimental results, the highest recognition accuracy of 88.44% was achieved by proposed recognition framework based on MFSC and PICNN, exceeding the performance of state-of-the-art deep learning algorithms such as CNN, VGGNet and Inception CNN (3.22%, 4.09% and 1.19%, respectively). These findings suggest that the newly developed silent speech approach holds promise to provide a more reliable communication channel, and the application scenery of speech recognition technology has been expanded at the same time.

Heart sound classification based on log Mel-frequency spectral coefficients features and convolutional neural networks

In view of the important role of heart sound signals in diagnosing and preventing congenital heart disease, a novel method about feature extraction and classification of heart sound signals was put forward in this study. Firstly, the heart sound signals were de-noised by using the wavelet algorithm. Subsequently, the improved duration-dependent hidden Markov model (DHMM) was used to segment the heart sound signal according to the heart cycle. Then, the dynamic frame length method was used to extract log Mel-frequency spectral coefficients (MFSC) features from the heart sound signal based on the heart cycle. Afterward, the convolution neural network (CNN) was used to classify the MFSC features. Finally, the majority voting algorithm was used to get the optimal classification results. In this paper, two-classification and multi-classification models were built. An accuracy of 93.89% for two-classification and an accuracy of 86.25% for multi-classification were achieved using the novel method.

Deep neural network architectures for dysarthric speech analysis and recognition

This paper investigates the ability of deep neural networks (DNNs) to improve the automatic recognition of dysarthric speech through the use of convolutional neural networks (CNNs) and long short-term memory (LSTM) neural networks. Dysarthria is one of the most common speech communication disorders associated with neurological impairments that can drastically reduce the intelligibility of speech. The aim of the present study is twofold. First, it compares three different input features for training and testing dysarthric speech recognition systems. These features are the mel-frequency cepstral coefficients (MFCCs), mel-frequency spectral coefficients (MFSCs), and the perceptual linear prediction features (PLPs). Second, the performance of the CNN- and LSTM-based architectures is compared against a state-of-the-art baseline system based on hidden Markov models (HMMs) and Gaussian mixture models (GMMs) to determine the best dysarthric speech recognizer. Experimental results show that the CNN-based system using perceptual linear prediction features provides a recognition rate that can reach 82%, which constitutes relative improvement of 11% and 32% when compared to the performance of LSTM- and GMM-HMM-based systems, respectively.

Estimation of ASR Parameterization for Interactive System

In this study, the authors explore the integration of speaker-independent automatic Amazigh speech recognition technology into interactive applications to extract data remotely from a distance database. Based on the combined interactive voice response (IVR) and automatic speech recognition (ASR) technologies, the authors built an interactive speech system to allow users to interact with the interactive system through voice commands. The hidden Markov models (HMMs), Gaussian mixture models (GMMs), and Mel frequency spectral coefficients (MFCCs) are used to develop a speech system based on the ten first Amazigh digits and six Amazigh words. The best-obtained performance is 89.64% by using 3 HMMs and 16 GMMs.

Machine Learning Evaluation for Music Genre Classification of Audio Signals

Music genre classification has its own popularity index in the present times. Machine learning can play an important role in the music streaming task. This research article proposes a machine learning based model for the classification of music genre. The evaluation of the proposed model is carried out while considering different music genres as in blues, metal, pop, country, classical, disco, jazz and hip-hop. Different audio features utilized in this study include MFCC (Mel Frequency Spectral Coefficients), Delta, Delta-Delta and temporal aspects for processing the data. The implementation of the proposed model has been done in the Python language. The results of the proposed model reveal an accuracy SVM accuracy of 95%. The proposed algorithm has been compared with existing algorithms and the proposed algorithm performs better than the existing ones in terms of accuracy.

Classification of heart sounds using fractional fourier transform based mel-frequency spectral coefficients and traditional classifiers

Heart sounds are a rich source of information for early diagnosis of cardiac pathologies. Distinguishing normal from abnormal heart sounds requires a specially trained clinician. Our goal is to develop a machine learning application that tackle the problem of heart sound classification. So we present a new processing and classification system for heart sounds. The automated diagnostic system is described in terms of its preprocessing, cardiac cycle segmentation, feature extraction, features reduction and classification stages. Conventional architectures will be used to identify abnormal heart sounds then the performances of the proposed systems will be compared. The conventional architectures include the following traditional classifiers: SVM, KNN and ensemble classifier (bagged Trees, subspace KNN and RUSBoosted tree). The proposed system is verified on the publicly available dataset of the heart sounds. The cross-validation and local hold out train-test methods are used to perform the experiments and obtain and compare the results. The proposed system showed potential for achieving excellent performance compared to previous methods on the same dataset with a score of 0.9200 at a sensitivity of 0.8735 and specificity of 0.9666 using a support vector machine classifier with cubic kernel. The details of the methodology and the results are presented and discussed.

Phonocardiogram Classification Using Deep Convolutional Neural Networks with Majority Vote Strategy

Most current automated phonocardiogram (PCG) classification methods are relied on PCG segmentation. It is universal to make use of the segmented PCG signals and then extract efficiency features for computer-aided auscultation or heart sound classification. However, the accurate segmentation of the fundamental heart sounds depends greatly on the quality of the heart sound signals. In addition these methods that heavily relied on segmentation algorithm considerably increase the computational burden. To solve above two issues, we have developed a novel approach to classify normal and abnormal cardiac diseases with un-segmented PCG signals. A deep Convolutional Neural Networks (DCNNs) method is proposed for recognizing normal and abnormal cardiac diseases. In the proposed method, one-dimensional heart sound signals are first converted into twodimensional feature maps which have three channels and each of them represents Mel-frequency spectral coefficients (MFSC) features including static, delta and delta–delta. These artificial images are then fed to the proposed DCNNs to train and evaluate normal and abnormal heart sound signals. We combined the method of majority vote strategy to finally obtain the category of PCG signals. Sensitivity (Se), Specificity (Sp) and Mean accuracy (MAcc) are used as the evaluation metrics. Results: Experiments demonstrated that our approach achieved a significant improvement, with the high Se, Sp, and MAcc of 92.73%, 96.90% and 94.81% respectively. The proposed method improves the MAcc by 5.63% compared with the best result in the CinC Challenge 2016. In addition, it has better robustness performance when applying for the long heart sounds. The proposed DCNNs-based method can achieve the best accuracy performance on recognizing normal and abnormal heart sounds without the preprocessing of segmental algorithm. It significantly improves the classification performance compared with the current state-of-art algorithm.

Speaker discrimination based on fuzzy fusion and feature reduction techniques

In this paper, we propose a research work on speaker discrimination using a multi-classifier fusion with focus on feature reduction effects. Speaker discrimination consists in the automatic distinction between two speakers using the vocal characteristics of their speeches. A number of features are extracted using Mel Frequency Spectral Coefficients and then reduced using Relative Speaker Characteristic (RSC) along with the Principal Components Analysis (PCA). Several classification methods are implemented to ensure the discrimination task. Since different classifiers are employed, two fusion algorithms at the decision level, referred to as Weighted Fusion and Fuzzy Fusion, are proposed to boost the classification performances. These algorithms are based on the weighting of the different classifiers outputs. Furthermore, the effects of speaker gender and feature reduction on the speaker discrimination task have been examined too. The evaluation of our approaches was conducted on a subset of Hub-4 Broadcast-News. The experimental results have shown that the speaker discrimination accuracy is improved by 5–15% using the (RSC–PCA) feature reduction. In addition, the proposed fusion methods recorded an improvement of about 10% compared to the individual scores of the classifiers. Finally, we noticed that the gender has an important impact on the discrimination performances.

Recognition of normal–abnormal phonocardiographic signals using deep convolutional neural networks and mel-frequency spectral coefficients

Intensive care unit patients are heavily monitored, and several clinically-relevant parameters are routinely extracted from high resolution signals. Objective: The goal of the 2016 PhysioNet/CinC Challenge was to encourage the creation of an intelligent system that fused information from different phonocardiographic signals to create a robust set of normal/abnormal signal detections. Approach: Deep convolutional neural networks and mel-frequency spectral coefficients were used for recognition of normal–abnormal phonocardiographic signals of the human heart. This technique was developed using the PhysioNet.org Heart Sound database and was submitted for scoring on the challenge test set. Main results: The current entry for the proposed approach obtained an overall score of 84.15% in the last phase of the challenge, which provided the sixth official score and differs from the best score of 86.02% by just 1.87%.

Modified DCTNet for audio signals classification

In this paper, we investigate DCTNet for audio signal classification. Its output feature is related to Cohen's class of time-frequency distributions. We introduce the use of adaptive DCTNet (A-DCTNet) for audio signals feature extraction. The A-DCTNet applies the idea of constant-Q transform, with its center frequencies of filterbanks geometrically spaced. The A-DCTNet is adaptive to different acoustic scales, and it can better capture low frequency acoustic information that is sensitive to human audio perception than features such as Mel-frequency spectral coefficients (MFSC). We use features extracted by the A-DCTNet as input for classifiers. Experimental results show that the A-DCTNet and Recurrent Neural Networks (RNN) achieve state-of-the-art performance in bird song classification rate, and improve artist identification accuracy in music data. They demonstrate A-DCTNet's applicability to signal processing problems.

The Research of Speech Recognition in Low SNR Based on GA-SVM

To improve recognizing rate and recognizing efficiency, the algorithm of double-threshold is adopted in the endpoint detection, Mel-frequency Spectral Coefficients is obtained as speech characteristic parameters, the SVM parameters, penalty factor and kernel function parameter, were optimized with genetic algorithm, then, an speech recognition model was established with GA-SVM method. The improved algorithm was simulated by matlab, the experiment shows that this algorithm can achieved good results in isolated word speech recognition.

Optimal Spectral Resolution in Speaker Authentication Application in Noisy Environment and Telephony

This article deals with the problem of speaker characterization, for which the principal interest is the improvement of the techniques of speaker authentication. For this purpose, we investigate the effect of spectral resolution in the speaker authentication performance. This investigation employs an approach based on the second order statistical measures using the Mel Frequency Spectral Coefficients (MFSC) and looks for the best spectral resolution. Experiments are done on microphonic and telephonic speech signals, with several MFSC dimensions and several types of noises. Results show the importance of the high spectral resolution in noisy environment and telephonic bandwidth, while the current research works have always favored the low resolution of 24 coefficients in such tasks. For example, we notice an improvement of about 11% on the identification score, since we increase the resolution from 24 to 48 MFSC, in the telephonic bandwidth.

Phonetic transition modeling for continuous speech recognition

Currently, most speech recognition architectures model the speech signal as a nonoverlapping sequence of phonetic segments. A set of phonetic models is created that attempt to capture the acoustic-phonetic properties of individual phones, but do not explicitly model the transition between phones. It is readily apparent, however, that these transitions contain important information about the identity of neighboring phones. While context-dependent phonetic modeling may capture some of this information, it is likely that more explicit models of phonetic transitions could offer performance improvements. In this talk, the use of phonetic transition models will be discussed within the context of summit, a segment-based continuous speech recognition system [Zue etal., ‘‘Acoustic Segmentation and Phonetic Classification in the summit Speech Recognition System,’’ Proc. ICASSP 89, pp. 389–392, Glasgow, Scotland (1989)]. The transition models use a feature vector based on Mel-frequency spectral coefficients (MFSC’s). The vector is created by concatenating multiple spectral averages on both sides of a transition. For example, in one configuration a total of eight averages were used which spanned a total time interval of 150 ms. In order to reduce the number of dimensions, a principal component analysis was performed. A set of diagonal Gaussian models is used to model the transitions. The models were tested by applying them to the N-Best sentence hypotheses from the recognition system. Each of the N-Best hypotheses is rescored using a linear combination (optimized on training data) of the segment and transition scores. Initial experiments have resulted in 10%–20% reductions in word error rates.