Filter Bank Features Research Articles

Analysis of Minimal Channel Electroencephalography for Wearable Brain–Computer Interface

Electroencephalography (EEG)-based brain—computer interface (BCI) is a non-invasive technology with potential in various healthcare applications, including stroke rehabilitation and neuro-feedback training. These applications typically require multi-channel EEG. However, setting up a multi-channel EEG headset is time-consuming, potentially resulting in patient reluctance to use the system despite its potential benefits. Therefore, we investigated the appropriate number of electrodes required for a successful BCI application in wearable devices using various numbers of EEG channels. EEG multi-frequency features were extracted using the “filter bank” feature extraction technique. A support vector machine (SVM) was used to classify a left/right-hand opening/closing motor imagery (MI) task. Nine electrodes around the center of the scalp (F3, Fz, F4, C3, Cz, C4, P3, Pz, and P4) provided high classification accuracy with a moderate setup time; hence, this system was selected as the minimal number of required channels. Spherical spline interpolation (SSI) was also applied to investigate the feasibility of generating EEG signals from limited channels on an EEG headset. We found classification accuracies of interpolated groups only, and combined interpolated and collected groups were significantly lower than the measured groups. The results indicate that SSI may not provide additional EEG data to improve classification accuracy of the collected minimal channels. The conclusion is that other techniques could be explored or a sufficient number of EEG channels must be collected without relying on generated data. Our proposed method, which uses a filter bank feature, session-dependent training, and the exploration of many groups of EEG channels, offers the possibility of developing a successful BCI application using minimal channels on an EEG device.

Graph signal processing is combined with deep learning for detection of damaged wind turbine blades

For early warning the damaged blade of wind turbines, an emission noise processing framework is proposed based on combination of Graph signal processing and Deep Learning. A microphone array is utilized to receive the noise emitted by the wind turbine blades. The weak abnormal signal from the damaged blade is enhanced by beamforming techniques. The enhanced signal is transformed into the graph domain by Graph Fourier Transform, from which the Mel filter bank features are extracted as inputs of a Multi-scale Feature Aggregation Conformer (MFA-Conformer) for damage detection. The MFA-Conformer combines Transformers and convolution neural networks (CNNs) to capture global and local features from the frequency or Graph domain. And the multi-stage aggregation strategy is utilized to exploit hierarchical context information. The reduction in the computational cost is achieved in the CNNs-based damage detection due to the real-valued features extracted from graph domain. The MFA-Conformer neural network is trained on the dataset which is created by applying data augmentation to the training samples. With the Mel filter bank features extracted from the frequency and graph domains, the MFA-Conformer neural network performs well in the five wind-farm data tests, with 2.55 % improvement in accuracy over the residual networks.

Leveraging Domain Features for Detecting Adversarial Attacks Against Deep Speech Recognition in Noise

In recent years, significant progress has been made in deep model-based automatic speech recognition (ASR), leading to its widespread deployment in the real world. At the same time, adversarial attacks against deep ASR systems are highly successful. Various methods have been proposed to defend ASR systems from these attacks. However, existing classification based methods focus on the design of deep learning models while lacking exploration of domain specific features. This work leverages filter bank-based features to better capture the characteristics of attacks for improved detection. Furthermore, the paper analyses the potentials of using speech and non-speech parts separately in detecting adversarial attacks. In the end, considering adverse environments where ASR systems may be deployed, we study the impact of acoustic noise of various types and signal-to-noise ratios. Extensive experiments show that the inverse filter bank features generally perform better in both clean and noisy environments, the detection is effective using either speech or non-speech part, and the acoustic noise can largely degrade the detection performance.

Identification of engine faults based on acoustic emission signals using a 1DCNN-ViT ensemble model

In view of the complexity of the engine mechanical structure and the diversity of faults, this paper presents a one-dimensional convolutional neural network (1DCNN)-vision transformer (ViT) ensemble model for identifying engine faults based on acoustic emission (AE) signals. The 1DCNN-ViT ensemble model combines 1DCNN and ViT. Firstly, AE signals of various faults are collected on the engine fault test rig. The dataset is constructed from its High-Mel Filterbank feature, which applies to AE signals. The proposed model has advantageous performance on this dataset. Secondly, the proposed model has a higher test accuracy than other new models. Finally, the fault data with different signal-to-noise ratios are input into the trained models, and the proposed model has better anti-noise performance. Overall, the proposed method can more accurately identify the AE signals of engine faults. It can be used as an effective method to diagnose engine faults.

Parameterization of voice onset for automatic assessment of Parkinson’s disease

Acoustic analysis of Parkinson’s disease (PD) usually focuses on sustained oscillations of the vocal folds; however, the voice onset (transition from the rest state to a saturation value) is often neglected. In this study, we investigated the parameterization of the voice onset for the objective evaluation of PD. 50 PD patients (25 females) and 50 healthy controls (25 females) performed the sustained phonation of the vowels /ah/, /ih/, and /uh/. Three experts listened to the recordings and assessed the dysarthria level according to a modified version of the Frenchay dysarthria assessment scale (mFDA). We extracted the voice onset automatically, considering the time needed for the amplitude envelope of the acoustic signal to go from 10% up to 90% of the maximum value. Then, we computed filter bank features from the voice onset and pitch, loudness, and perturbation features from the sustained phonation. We performed automatic classification and regression analyses using support vector machines. We obtained a classification accuracy of 89% (AUC: 0.93) when we used feature importance analysis to reduce the feature set. In addition, the regression analysis showed a significant correlation (r: 0.583, p-value < 0.001) between the acoustic features and the mFDA scale. These results show the suitability of the voice onset for objective clinical evaluation of PD.

Signal identification based on modified filter bank feature and generalized regression neural network for optical fiber perimeter sensing

• Signal recognition method based on the generalized regression neural network (GRNN). • Feature extraction algorithm based on the modified filter bank (FB) feature. • Endpoint detection algorithm based on the short-time energy. • Little parameter adjustment and human influence in the training process. • High identification rate, short recognition time and broad application prospects. A method based on the modified filter bank (FB) feature and a generalized regression neural network (GRNN) is proposed to raise the accuracy and real-time performance of signal recognition in optical fiber perimeter sensing systems. The FB feature is modified by adding root mean square information of the signal power under the Mel filter bank. Four kinds of sensing signals under three types of weather conditions are obtained experimentally by a fenced perimeter sensing system based on an in-line Sagnac interferometer. The endpoint detection algorithm based on short-term energy is performed on the sensing signals to obtain the effective signal segments. Then the modified FB features of signal segments are extracted and randomly divided into training and testing samples. The optimal GRNN classifier model is generated by performing a 10-fold cross-validation on the training samples and used to recognize the testing samples. The average accuracy can reach 98.22 % and average recognition time is only about 0.07 s. This method is expected to meet the requirements for real-time and accuracy in practical application of optical fiber perimeter sensing signal recognition.

Multirate Audiometric Filter Bank for Hearing Aid Devices.

The frequency-dependent nature of hearing loss poses many challenges for hearing aid design. In order to compensate for a hearing aid user's unique hearing loss pattern, an input signal often needs to be separated into frequency bands, or channels, through a process called sub-band decomposition. In this paper, we present a real-time filter bank for hearing aids. Our filter bank features 10 channels uniformly distributed on the logarithmic scale, located at the standard audiometric frequencies used for the characterization and fitting of hearing aids. We obtained filters with very narrow passbands in the lower frequencies by employing multi-rate signal processing. Our filter bank offers a 9.1× reduction in complexity as compared to conventional signal processing. We implemented our filter bank on Open Speech Platform, an open-source hearing aid, and confirmed real-time operation.

Text-independent speaker recognition based on adaptive course learning loss and deep residual network

Text-independent speaker recognition is widely used in identity recognition that has a wide spectrum of applications, such as criminal investigation, payment certification, and interest-based customer services. In order to improve the recognition ability of log filter bank feature vectors, a method of text-independent speaker recognition based on deep residual networks model was proposed in this paper. The deep residual network was composed of a residual network (ResNet) and a convolutional attention statistics pooling (CASP) layer. The CASP layer could aggregate frame-level features from the ResNet into an utterance-level features. Extracting speech features for each speaker using deep residual networks was a promising direction to explore, and a straightforward solution was to train the discriminative feature extraction network by using a margin-based loss function. However, a margin-based loss function often has certain limitations, such as the margins between different categories were set to be the same and fixed. Thus, we used an adaptive curriculum learning loss (ACLL) to address the problem and introduce two different margin-based losses for this problem, i.e., AM-Softmax and AAM-Softmax. The proposed method was applied to a large-scale VoxCeleb2 dataset for extensive text-independent speaker recognition experiments, and average equal error rate (EER) could achieve 1.76% on VoxCeleb1 test dataset, 1.91% on VoxCeleb1-E test dataset, and 3.24% on VoxCeleb1-H test dataset. Compared with related speaker recognition methods, EER was improved by 1.11% on VoxCeleb1 test dataset, 1.04% on VoxCeleb1-E test dataset, and 1.69% on VoxCeleb1-H test dataset.

Diagnosing the Stage of COVID-19 using Machine Learning on Breath Sounds

With rapidly increasing COVID-19 cases, patients with mildand moderate symptoms are being asked to home-isolate themselvesto save hospital resources for more severe patients.Such patients have been asked to self-monitor themselves andseek medical attention if their condition worsens. COVID-19 affects the respiratory system and home-isolated patientsmust monitor their lung condition continuously before it quicklydeteriorates. But this is difficult to monitor by oneself, and thepatient may not notice his worsening lung condition before itis too late. A machine-learning based approach is proposedto monitor lung condition by analyzing the breath sounds ofa patient for respiratory sounds like wheezes, crackles andtachypnea, which in turn can identify the stage of COVID-19. Data from a respiratory sound database with recordingsfrom 226 patients was split into 6898 respiratory cycles andpre-processed. In this paper, two approaches are evaluated.The first approach is demonstrated using Google Cloud AutoMLwith the recordings of respiratory cycles which wereconverted to spectrograms to train the model. In the secondapproach, Log Mel filter-bank features were extracted fromthe breath sounds and used to train multiple CNN models tohierarchically classify breath sounds. This ensemble-learningwith hierarchical-model approach achieved a better accuracyof 78.12%. This model can be integrated with a mobile applicationto record and analyze breath-sounds. This will enablethe patient to admit himself sooner if he is progressing to asevere stage of COVID-19.

NANO-studio, the design environment of filter banks implemented in standard CMOS technology

The paper presents a method of optimizing frequency characteristics of filter banks in terms of their implementation in digital CMOS technologies in nanoscale. Usability of such filters is demonstrated by frequency-interleaved (FI) analog-to-digital converters (ADC). An analysis filter present in these converters was designed in switched-current technique. However, due to huge technological pitch of standard digital CMOS process in nanoscale, its characteristics substantially deviate from the required ones. NANO-studio environment presented in the paper allows adjustment, with transistor channel sizes as optimization parameters. The same environment is used at designing a digital synthesis filter, whereas optimization parameters are input and output conductances, gyration transconductances and capacitances of a prototype circuit. Transition between analog s and digital z domains is done by means of bilinear transformation. Assuming a lossless gyrator-capacitor (gC) multiport network as a prototype circuit, both for analysis and synthesis filter banks in FI ADC, is an implementation of the strategy to design filters with low sensitivity to parameter changes. An additional advantage is designing the synthesis filter as stable infinite impulse response (IIR) instead of commonly used finite impulse response (FIR) filters. It provides several dozen-fold saving in the number of applied multipliers.. The analysis and synthesis filters in FI ADC are implemented as filter pairs. An additional example of three-filter bank demonstrates versatility of NANO-studio software.

Acoustic feature based unsupervised approach of heart sound event detection

This paper represents an unsupervised approach to detect the positions of S1, S2 heart sound events in a Phonocardiogram (PCG) recording. Insufficiency of correctly annotated heart sound database drives us to investigate unsupervised techniques. Gammatone filter bank features are used to characterize the spectral pattern of fundamental heart sound events from noise contaminated PCG data. An unsupervised spectral clustering technique is employed for segmentation of S1/S2 and non-S1/S2 heart sound events. A Feature winning score is computed to identify the S1/S2 and non-S1/S2 frames. Finally, time based threshold is applied to detect the accurate positions of S1 and S2 heart sounds. The performance of spectral clustering is compared with other clustering methods. The proposed method offers a maximum F1-score of 98% and 92.5% for normal and abnormal PCG data respectively on 2016 PhysioNet/CinC challenge dataset. The heart sound annotation algorithm provided by PhysioNet has been used as the ground truth after hand correction.

Motor imagery classification using geodesic filtering common spatial pattern and filter-bank feature weighted support vector machine.

In recent years, Brain Computer Interface (BCI) based on motor imagery has been widely used in the fields of medicine, active safe systems for automobiles, entertainment, and so on. Motor imagery relevant electroencephalogram (EEG) signals are weak, nonlinear, and susceptible to interference. As a feature extraction method for motor imagery, Common Spatial Pattern (CSP) has been proven to be very effective. However, its effectiveness depends heavily on the choice of frequency bands, and Euclidean space cannot effectively describe the inner relationship. To solve these problems, a classification approach for motor imagery using the Geodesic Filtering Common Spatial Pattern (GFCSP) and filter-bank Feature Weighted Support Vector Machine (FWSVM) is presented. First, GFCSP based on the Riemannian manifold is proposed, in which the extracted covariance features are spatially filtered in Riemannian tangent space, and the average covariance matrix is replaced by Riemannian mean in CSP. Second, filter-bank FWSVM with a feature weighted matrix is proposed. EEG signals are filtered into 8-12 Hz, 12-16 Hz, 18-22 Hz, 22-26 Hz, and a wide band of 8-24 Hz, and GFCSP features of these filtered signals are extracted. A feature weighted matrix is calculated using mutual information and the Pearson correlation coefficient from these features and class information. Then, the Support Vector Machine (SVM) is used for classification with the feature weighted matrix. Finally, the proposed method is validated on the dataset IVa in BCI competition III. Classification accuracies of the five subjects are 92.31%, 99.03%, 80.36%, 96.30%, and 97.67%, which demonstrate the effectiveness of our proposed method.

Audio Replay Spoof Attack Detection by Joint Segment-Based Linear Filter Bank Feature Extraction and Attention-Enhanced DenseNet-BiLSTM Network

Most automatic speaker verification (ASV) systems are vulnerable to various spoofing attacks. In recent years, there have been many methods were proposed for detecting spoofing attacks in ASV, and significant progress has been made. However, current methods have shown little improvements in replay spoof attack detection as they lack a more suitable model for replay detection. To address this issue, in this article, we propose a novel model based on attention-enhanced DenseNet-BiLSTM network and segment-based linear filter bank features. First, silent segments are selected from each speech signal by using a short-term zero-crossing rate and energy. If the total duration of silent segments only contains a very limited amount of data, the decaying tails will be selected instead. Second, the linear filter bank features are extracted from the selected segments in the relatively high-frequency domain. Finally, an attention-enhanced DenseNet-BiLSTM architecture which can avoid the problems of overfitting is built. To validate this model, we used two datasets, including BTAS2016 and ASVspoof2017. Experiments show that using the attention-enhanced DenseNet-BiLSTM model with the segment-based linear filter bank feature achieves the best performance. Compared with the baseline system based on constant Q cepstral coefficient and Gaussian mixture model (GMM), the proposed model can produce a relative improvement of 91.68% and 74.04% on the two data sets respectively.

A Hybrid Approach for Single Channel Speech Enhancement using Deep Neural Network and Harmonic Regeneration Noise Reduction

This paper presents a hybrid approach for single channel speech enhancement using deep neural network (DNN) and harmonic regeneration noise reduction (HRNR). The DNN was used as a supervised algorithm to predict new target mask such as constrained Wiener Filter (cWF) target mask from noisy mixture signal that was transformed into gammatone filter bank features. Meanwhile, HRNR algorithm was applied in the post-filtering strategy to eliminate residual noise. The DNN algorithm is an emerging supervised speech enhancement to overcome heavy nonstationary noise and low signal-to-noise ratio (SNR) issues. To validate the proposed algorithm with new target mask, 600 Malay utterances combining male and female speakers were used in a training session while 120 Malay utterances were used in a prediction session. The short time objective intelligibility (STOI) and perceptual evaluation of speech quality (PESQ) scores were calculated as the performance metrics. In this work, the proposed target mask outperformed other baseline target masks. Thus, PESQ and STOI scores for the hybrid speech enhancement algorithm is 1.17 and 0.79, respectively, at - 5 dB babble noise SNR.

Modeling the onset advantage in musical instrument recognition.

Sound onsets provide particularly valuable cues for musical instrument identification by human listeners. It has remained unclear whether this onset advantage is due to enhanced perceptual encoding or the richness of acoustical information during onsets. Here this issue was approached by modeling a recent study on instrument identification from tone excerpts [Siedenburg. (2019). J. Acoust. Soc. Am. 145(2), 1078-1087]. A simple Hidden Markov Model classifier with separable Gabor filterbank features simulated human performance and replicated the onset advantage observed previously for human listeners. These results provide evidence that the onset advantage may be driven by the distinct acoustic qualities of onsets.

Spoken Language Understanding of Human-Machine Conversations for Language Learning Applications

Spoken language understanding (SLU) in human machine conversational systems is the process of interpreting the semantic meaning conveyed by a user’s spoken utterance. Traditional SLU approaches transform the word string transcribed by an automatic speech recognition (ASR) system into a semantic label that determines the machine’s subsequent response. However, the robustness of SLU results can suffer in the context of a human-machine conversation-based language learning system due to the presence of ambient noise, heavily accented pronunciation, ungrammatical utterances, etc. To address these issues, this paper proposes an end-to-end (E2E) modeling approach for SLU and evaluates the semantic labeling performance of a bidirectional LSTM-RNN with input at three different levels: acoustic (filterbank features), phonetic (subphone posteriorgrams), and lexical (ASR hypotheses). Experimental results for spoken responses collected in a dialog application designed for English learners to practice job interviewing skills show that multi-level BLSTM-RNNs can utilize complementary information from the three different levels to improve the semantic labeling performance. An analysis of results on OOV utterances, which can be common in a conversation-based dialog system, also indicates that using subphone posteriorgrams outperforms ASR hypotheses and incorporating the lower-level features for semantic labeling can be advantageous to improving the final SLU performance.

Convolutional Neural Network-based Speech Enhancement for Cochlear Implant Recipients.

Attempts to develop speech enhancement algorithms with improved speech intelligibility for cochlear implant (CI) users have met with limited success. To improve speech enhancement methods for CI users, we propose to perform speech enhancement in a cochlear filter-bank feature space, a feature-set specifically designed for CI users based on CI auditory stimuli. We leverage a convolutional neural network (CNN) to extract both stationary and non-stationary components of environmental acoustics and speech. We propose three CNN architectures: (1) vanilla CNN that directly generates the enhanced signal; (2) spectral-subtraction-style CNN (SS-CNN) that first predicts noise and then generates the enhanced signal by subtracting noise from the noisy signal; (3) Wiener-style CNN (Wiener-CNN) that generates an optimal mask for suppressing noise. An important problem of the proposed networks is that they introduce considerable delays, which limits their real-time application for CI users. To address this, this study also considers causal variations of these networks. Our experiments show that the proposed networks (both causal and non-causal forms) achieve significant improvement over existing baseline systems. We also found that causal Wiener-CNN outperforms other networks, and leads to the best overall envelope coefficient measure (ECM). The proposed algorithms represent a viable option for implementation on the CCi-MOBILE research platform as a pre-processor for CI users in naturalistic environments.

Mel scaled M-band wavelet filter bank for speech recognition

A Mel scaled M-band wavelet filter bank structure is used to extract the robust acoustic feature for speech recognition application. The proposed filter bank can provide flexibility of frequency partition that decomposes the speech signal into the M-frequency band. To estimate the difference between Mel scaled M-band wavelet and dyadic wavelet filter bank, relative bandwidth deviation (RBD) and root mean square bandwidth deviation (RMSBD) with respect to baseline (Mel filter bank bandwidth) is calculated. Proposed filter bank gives 40.90 and 49.84% reduction for RBD and RMSBD respectively, over 24-dyadic wavelet filter bank. Feature extraction from the proposed filter bank using AMUAV corpus shows an improvement in terms of word recognition accuracy (WRA) at all SNR range (20 dB to 0 dB) over baseline (MFCC) features. For AMUAV corpus, the proposed feature shows the maximum improvement in WRA of 3.93% over baseline features and 3.90% over dyadic wavelet filter bank features. When applied to the VidTIMIT corpus, proposed features show the maximum improvement in WRA of 1.64% over baseline features and 4.43% over dyadic features.

FMLLR Speaker Normalization With i-Vector: In Pseudo-FMLLR and Distillation Framework

When an automatic speech recognition (ASR) system is deployed for real-world applications, it often receives only one utterance at a time for decoding. This single utterance could be of short duration depending on the ASR task. In these cases, robust estimation of speaker normalizing methods like feature-space maximum likelihood linear regression (FMLLR) and i -vectors may not be feasible. In this paper, we propose two unsupervised speaker normalization techniques—one at feature level and other at model level of acoustic modeling—to overcome the drawbacks of FMLLR and i -vectors in real-time scenarios. At feature level, we propose the use of deep neural networks (DNN) to generate pseudo-FMLLR features from time-synchronous pair of filterbank and FMLLR features. These pseudo-FMLLR features can then be used for DNN acoustic model training and decoding. At model level, we propose a generalized distillation framework, where a teacher DNN trained on FMLLR features guides the training and optimization of a student DNN trained on filterbank features. In both the proposed methods, the ambiguity in choosing the speaker-specific FMLLR transform can be reduced by augmenting i -vectors to the input filterbank features. Experiments conducted on 33-h and 110-h subsets of Switchboard corpus show that the proposed methods provide significant gains over DNNs trained on FMLLR, i -vector appended FMLLR, filterbank and i -vector appended filterbank features, in real-time scenario.

Texture Classification in Extreme Scale Variations Using GANet.

Research in texture recognition often concentrates on recognizing textures with intraclass variations, such as illumination, rotation, viewpoint, and small-scale changes. In contrast, in real-world applications, a change in scale can have a dramatic impact on texture appearance to the point of changing completely from one texture category to another. As a result, texture variations due to changes in scale are among the hardest to handle. In this paper, we conduct the first study of classifying textures with extreme variations in scale. To address this issue, we first propose and then reduce scale proposals on the basis of dominant texture patterns. Motivated by the challenges posed by this problem, we propose a new GANet network where we use a genetic algorithm to change the filters in the hidden layers during network training in order to promote the learning of more informative semantic texture patterns. Finally, we adopt a Fisher vector pooling of a convolutional neural network filter bank feature encoder for global texture representation. Because extreme scale variations are not necessarily present in most standard texture databases, to support the proposed extreme-scale aspects of texture understanding, we are developing a new dataset, the extreme scale variation textures (ESVaT), to test the performance of our framework. It is demonstrated that the proposed framework significantly outperforms the gold-standard texture features by more than 10% on ESVaT. We also test the performance of our proposed approach on the KTHTIPS2b and OS datasets and a further dataset synthetically derived from Forrest, showing the superior performance compared with the state-of-the-art.