Linear Frequency Cepstral Coefficients Research Articles

Cepstral coefficients effectiveness for gunshot classifying

This paper analyses the efficiency of various frequency cepstral coefficients (FCC) in a non-speech application, specifically in classifying acoustic impulse events-gunshots. There are various methods for such event identification available. The majority of these methods are based on time or frequency domain algorithms. However, both of these domains have their limitations and disadvantages. In this article, an FCC, combining the advantages of both frequency and time domains, is presented and analyzed. These originally speech features showed potential not only in speech-related applications but also in other acoustic applications. The comparison of the classification efficiency based on features obtained using four different FCC, namely mel-FCC (MFCC), inverse mel-frequency cepstral coefficients (IMFCC), linear-frequency cepstral coefficients (LFCC), and gammatone-frequency cepstral coefficients (GTCC) is presented. An optimal frame length for an FCC calculation is also explored. Various gunshots from short guns and rifle guns of different calibers and multiple acoustic impulse events, similar to the gunshots, to represent false alarms are used. More than 600 acoustic events records have been acquired and used for training and validation of two designed classifiers, support vector machine, and neural network. Accuracy, recall and Matthew’s correlation coefficient measure the classification success rate. The results reveal the superiority of GFCC to other analyzed methods.

A convolutional neural network with image and numerical data to improve farming of edible crickets as a source of food-A decision support system.

Crickets (Gryllus bimaculatus) produce sounds as a natural means to communicate and convey various behaviors and activities, including mating, feeding, aggression, distress, and more. These vocalizations are intricately linked to prevailing environmental conditions such as temperature and humidity. By accurately monitoring, identifying, and appropriately addressing these behaviors and activities, the farming and production of crickets can be enhanced. This research implemented a decision support system that leverages machine learning (ML) algorithms to decode and classify cricket songs, along with their associated key weather variables (temperature and humidity). Videos capturing cricket behavior and weather variables were recorded. From these videos, sound signals were extracted and classified such as calling, aggression, and courtship. Numerical and image features were extracted from the sound signals and combined with the weather variables. The extracted numerical features, i.e., Mel-Frequency Cepstral Coefficients (MFCC), Linear Frequency Cepstral Coefficients, and chroma, were used to train shallow (support vector machine, k-nearest neighbors, and random forest (RF)) ML algorithms. While image features, i.e., spectrograms, were used to train different state-of-the-art deep ML models, i,e., convolutional neural network architectures (ResNet152V2, VGG16, and EfficientNetB4). In the deep ML category, ResNet152V2 had the best accuracy of 99.42%. The RF algorithm had the best accuracy of 95.63% in the shallow ML category when trained with a combination of MFCC+chroma and after feature selection. In descending order of importance, the top 6 ranked features in the RF algorithm were, namely humidity, temperature, C#, mfcc11, mfcc10, and D. From the selected features, it is notable that temperature and humidity are necessary for growth and metabolic activities in insects. Moreover, the songs produced by certain cricket species naturally align to musical tones such as C# and D as ranked by the algorithm. Using this knowledge, a decision support system was built to guide farmers about the optimal temperature and humidity ranges and interpret the songs (calling, aggression, and courtship) in relation to weather variables. With this information, farmers can put in place suitable measures such as temperature regulation, humidity control, addressing aggressors, and other relevant interventions to minimize or eliminate losses and enhance cricket production.

Complementary regional energy features for spoofed speech detection

Automatic speaker verification systems are found to be vulnerable to spoof attacks such as voice conversion, text-to-speech, and replayed speech. As the security of biometric systems is vital, many countermeasures have been developed for spoofed speech detection. To satisfy the recent developments on speech synthesis, publicly available datasets became more and more challenging (e.g., ASVspoof 2019 and 2021 datasets). A variety of replay attack configurations were also considered in those datasets, as they do not require expertise, hence easily performed. This work utilizes regional energy features, which are experimentally proven to be more effective than the traditional frame-based energy features. The proposed energy features are independent from the utterance length and are extracted over nonoverlapping time-frequency regions of the magnitude spectrum. Different configurations are considered in the experiments to verify the regional energy features’ contribution to the performance. First, light convolutional neural network – long short-term memory (LCNN – LSTM) model with linear frequency cepstral coefficients is used to determine the optimal number of regional energy features. Then, SE-Res2Net model with log power spectrogram features is used, which achieved comparable results to the state-of-the-art for ASVspoof 2019 logical access condition. Physical access condition from ASVspoof 2019 dataset, logical access and deep fake conditions from ASVspoof 2021 dataset are also used in the experiments. The regional energy features achieved improvements for all conditions with almost no additional computational or memory loads (less than 1% increase in the model size for SE-Res2Net). The main advantages of the regional energy features can be summarized as i) capturing nonspeech segments, ii) extracting band-limited information. Both aspects are found to be discriminative for spoofed speech detection.

Application of Teager Energy Operator on Linear and Mel Scales for Whispered Speech Recognition

This paper presents experimental results on whispered speech recognition based on Teager Energy Operator for linear and mel cepstral coefficients including the Cepstral Mean Subtraction normalization technique. The feature vectors taken into consideration are Linear Frequency Cepstral Coefficients, Teager Energy based Linear Frequency Cepstral Coefficients, Mel Frequency Cepstral Coefficients and Teager Energy based Mel Frequency Cepstral Coefficients. A speaker dependent scenario is used. For the recognition process, Dynamic Time Warping and Hidden Markov Models methods are applied. Results show a respectable improvement in whispered speech recognition as achieved by using the Teager Energy Operator with Cepstral Mean Subtraction.

An Improved AlexNet Model and Cepstral Coefficient-Based Classification of Autism Using EEG.

Autism is a neurodevelopmental disorder that cannot be completely cured, but early intervention during childhood can improve outcomes. Identifying autism spectrum disorder (ASD) has relied on subjective detection methods that involve questionnaires, medical professionals, and therapists and are subject to observer variability. The need for early diagnosis and the limitations of subjective detection methods has led researchers to explore machine learning-based approaches, such as Random Forests, K-Nearest Neighbors, Naive Bayes, and Support Vector Machines, to predict ASD meltdowns. In recent years, deep learning techniques have gained traction for early ASD detection. This study evaluates the performance of various deep learning networks, including AlexNet, VGG16, and ResNet50, using 5 cepstral coefficient features for ASD detection. The main contributions of this study are the utilization of Cepstral Coefficients in the processing stage to construct spectrograms and the modification of the AlexNet architecture for precise classification. Experimental observations indicate that the AlexNet with Linear Frequency Cepstral Coefficients (LFCC) yields the highest accuracy of 85.1%, while a customized AlexNet with LFCC achieves 90% accuracy.

Effective preservation of higher-frequency contents in the context of short utterance based children’s speaker verification system

Developing an automatic speaker verification (ASV) system for children is extremely challenging due to the unavailability of children’s speech corpora. The challenges are further exacerbated in the case of short utterances. Voice-based biometric systems require adequate amount of speech data for enrollment and verification; otherwise the performance considerably degrades. In this paper, we have focussed on data paucity and preserving the higher-frequency contents in order to enhance the performance of a short-utterance based children’s speaker verification system. To deal with data scarcity, several out-of-domain data augmentation techniques have been utilized. Since the out-of-domain data used is from adult speakers which are acoustically very different from children’s speech, we have resorted to techniques like prosody modification, formant modification and voice-conversion in order to render it acoustically similar to children’s speech prior to augmentation. This helps in not only increasing the amount of training data but also in effectively capturing the missing target attributes. A relative improvement of 33.6% in equal error rate (EER) with respect to the baseline system trained solely on child data-set is achieved when the proposed data augmentation technique is applied. Further to that, for the preservation of the higher-frequency contents, we have resorted to concatenation of the classical Mel-frequency cepstral coefficients (MFCC) features with the linear-frequency cepstral coefficient (LFCC) or with the inverse-Mel-frequency cepstral coefficient (IMFCC) features. The use of Mel-filter-bank leads to poor resolution of higher-frequency components. On the other hand, linear- or inverse-Mel-filter-banks yield better resolution of higher-frequency components. Moreover, MFCC and IMFCC features exhibit low canonical correlation. Consequently, the frame-level concatenation of MFCC and LFCC or IMFCC features leads to better resolution of both lower- as well as higher-frequency components. Therefore, the EER considerably reduces when either LFCC features or IMFCC features are concatenated with MFCC features. The EER for the full test set shows a relative reduction of 10.56% (with respect to the EER for the MFCC features) when IMFCC features are concatenated with the MFCC features. This novel approach of incorporating data augmentation followed by frame-level feature concatenation helps in achieving an overall reduction of 40.6% in EER.

The soundscape of swarming: Proof of concept for a noninvasive acoustic species identification of swarming Myotis bats.

Bats emit echolocation calls to orientate in their predominantly dark environment. Recording of species-specific calls can facilitate species identification, especially when mist netting is not feasible. However, some taxa, such as Myotis bats can be hard to distinguish acoustically. In crowded situations where calls of many individuals overlap, the subtle differences between species are additionally attenuated. Here, we sought to noninvasively study the phenology of Myotis bats during autumn swarming at a prominent hibernaculum. To do so, we recorded sequences of overlapping echolocation calls (N= 564) during nights of high swarming activity and extracted spectral parameters (peak frequency, start frequency, spectral centroid) and linear frequency cepstral coefficients (LFCCs), which additionally encompass the timbre (vocal "color") of calls. We used this parameter combination in a stepwise discriminant function analysis (DFA) to classify the call sequences to species level. A set of previously identified call sequences of single flying Myotis daubentonii and Myotis nattereri, the most common species at our study site, functioned as a training set for the DFA. 90.2% of the call sequences could be assigned to either M. daubentonii or M. nattereri, indicating the predominantly swarming species at the time of recording. We verified our results by correctly classifying the second set of previously identified call sequences with an accuracy of 100%. In addition, our acoustic species classification corresponds well to the existing knowledge on swarming phenology at the hibernaculum. Moreover, we successfully classified call sequences from a different hibernaculum to species level and verified our classification results by capturing swarming bats while we recorded them. Our findings provide a proof of concept for a new noninvasive acoustic monitoring technique that analyses "swarming soundscapes" by combining classical acoustic parameters and LFCCs, instead of analyzing single calls. Our approach for species identification is especially beneficial in situations with multiple calling individuals, such as autumn swarming.

On significance of constant-Q transform for pop noise detection

Liveness detection has emerged as an important research issue for many biometrics, such as face, iris, hand geometry, etc. and significant research efforts are reported in the literature. However, less emphasis is given to liveness detection for voice biometrics or Automatic Speaker Verification (ASV). Voice Liveness Detection (VLD) can be a potential technique to detect spoofing attacks in ASV system. Presence of pop noise in the speech signal of live speaker provides the discriminative acoustic cue to distinguish between genuine vs. spoofed speech in the framework of VLD. Pop noise comes out as a burst at the lips, which is captured by the ASV system (since the speaker and microphone are close enough), indicating the liveness of the speaker and provides the basis of VLD. In this paper, we present the Constant-Q Transform (CQT) -based approach over the traditional Short-Time Fourier Transform (STFT) -based algorithm (baseline). With respect to Heisenberg’s uncertainty principle in signal processing framework, the CQT has variable spectro-temporal resolution, in particular, better frequency resolution for low frequency region and better temporal resolution for high frequency region, which can be effectively utilized to identify the low frequency characteristics of pop noise. We have also compared proposed algorithm with cepstral features, namely, Linear Frequency Cepstral Coefficients (LFCC) and Constant-Q Cepstral Coefficients. The experiments are performed on recently released POp noise COrpus (POCO) dataset with various statistical, discriminative, and deep learning-based classifiers, namely, Gaussian Mixture Model (GMM), Support Vector Machine (SVM), Convolutional Neural Networks (CNN), Light-CNN (LCNN), and Residual Network (ResNet), respectively. The significant improvement in performance, in particular, an absolute improvement of 14.23% and 10.95% in terms of percentage classification accuracy on development and evaluation set, respectively, is obtained for the proposed CQT-based algorithm along with SVM classifier, over the STFT-SVM (baseline) system. Similar trend of the performance improvement is observed for the GMM, CNN, LCNN, and ResNet classifiers for the proposed CQT-based algorithm vs. traditional STFT-based algorithm. The analysis is further extended by simulating the replay mechanism (in the standard framework of ASVSpoof-2019 PA challenge dataset) on the subset of POCO dataset in order to observe the effect of room acoustics onto the performance of the VLD system. By embedding the moderate simulated replay mechanism in POCO dataset, we obtained the percentage classification accuracy of 97.82% on evaluation set.

Implementation of Multiple Feature Selection Algorithms for Speech Spoofing Detection

The ASVspoof challenge sequences were proposed to lead the research in anti-spoofing to a new level for automatic speaker verification (ASV). It’s verified that constant Q cepstral coefficients (CQCC) processes speech in variable frequencies with adjustable resolution and outperforms the other generally used features and Linear Frequency Cepstral Coefficient (LFCC) is used in high-frequency areas. The feature selection algorithm is offered to decrease computational complexity and overfitting for spoofed utterance detection. Precisely, there’s a demand for feature selection algorithms that are computationally effective and sensitive to feature interactions so that useful features aren’t falsely excluded during the ranking process. We experiment on the ASVspoof 2019 challenge for the assessment of spoofing countermeasures. After the evaluation of our given algorithms and data gives us an equal error rate (EER) and tandem discovery cost function (t-DCF) values. Experimental results on ASVspoof 2019 physical access referring to multiple feature selection approaches show a breakthrough compared to the baseline.

Automatic discrimination between front and back ensemble locations in HRTF-convolved binaural recordings of music

One of the greatest challenges in the development of binaural machine audition systems is the disambiguation between front and back audio sources, particularly in complex spatial audio scenes. The goal of this work was to develop a method for discriminating between front and back located ensembles in binaural recordings of music. To this end, 22, 496 binaural excerpts, representing either front or back located ensembles, were synthesized by convolving multi-track music recordings with 74 sets of head-related transfer functions (HRTF). The discrimination method was developed based on the traditional approach, involving hand-engineering of features, as well as using a deep learning technique incorporating the convolutional neural network (CNN). According to the results obtained under HRTF-dependent test conditions, CNN showed a very high discrimination accuracy (99.4%), slightly outperforming the traditional method. However, under the HRTF-independent test scenario, CNN performed worse than the traditional algorithm, highlighting the importance of testing the algorithms under HRTF-independent conditions and indicating that the traditional method might be more generalizable than CNN. A minimum of 20 HRTFs are required to achieve a satisfactory generalization performance for the traditional algorithm and 30 HRTFs for CNN. The minimum duration of audio excerpts required by both the traditional and CNN-based methods was assessed as 3 s. Feature importance analysis, based on a gradient attribution mapping technique, revealed that for both the traditional and the deep learning methods, a frequency band between 5 and 6 kHz is particularly important in terms of the discrimination between front and back ensemble locations. Linear-frequency cepstral coefficients, interaural level differences, and audio bandwidth were identified as the key descriptors facilitating the discrimination process using the traditional approach.

Voice Spoofing Countermeasure Based on Spectral Features to Detect Synthetic Attacks Through LSTM

With the growing number of voice-controlled devices, it is necessary to address the potential vulnerabilities of Automatic Speaker Verification (ASV) against voice spoofing attacks such as Physical Access (PA) and Logical Access (LA) attacks. To improve the reliability of ASV systems, researchers have developed various voice spoofing countermeasures. However, it is hard for the voice anti-spoofing systems to effectively detect the synthetic speech attacks that are generated through powerful spoofing algorithms and have quite different statistical distributions. More importantly, the speedy improvement of voice spoofing structures is producing the most effective attacks that make ASV structures greater vulnerable to stumble on those voice spoofing assaults. In this paper, we proposed a unique voice spoofing countermeasure which is successful to hit upon the LA attacks (i.e., artificial speech and transformed speech) and classify the spoofing structures by the usage of Long Short-Term Reminiscence (LSTM). The novel set of spectral features i.e., Mel-Frequency Cepstral Coefficients (MFCC), Gammatone Cepstral Coefficients (GTCC), and spectral centroid are capable to seize maximum alterations present in the cloned audio. The proposed system achieved remarkable accuracy of 98.93%, precision of 100%, recall of 92.32%, F1-score of 96.01%, and an Equal Error Rate (EER) of 1.30%. Our method achieved 8.5% and 7.02% smaller EER than the baseline methods such as Constant-Q Cepstral Coefficients (CQCC) using Gaussian Mixture Model (GMM) and Linear Frequency Cepstral Coefficients (LFCC) using GMM, respectively. We evaluated the performance of the proposed system on the standard dataset i.e., ASVspoof2019 LA. Experimental results and comparative analysis with other existing state-of-the-art methods illustrate that our method is reliable and effective to be used for the detection of voice spoofing attacks.

The speech synthesis detection algorithm based on cepstral coefficients and convolutional neural network

The existing approaches to detecting synthesized speech, based on the current issues of synthesizing voice sequences, are considered. The stages of the algorithm for detecting spoofing attacks on voice biometric systems are described, and its final workflow is presented. The research focuses mainly on detecting synthesized speech, as it is the most dangerous type of attacks. The authors designed a software application for an experimental study, present its structure and propose the detection synthesized speech algorithm. This algorithm uses mel-frequency and constant Q cepstral coefficients to extract speech features. A Gaussian mixture model is used to construct a user model. Convolutional neural network was chosen as a classifier to determine the voice’s authenticity. Two basic methods for combating spoofing attacks, proposed by the authors of the ASVspoof2019 competition, were selected for making comparisons. One of these methods involved using linear frequency cepstral coefficients as speech features, while the other method used constant Q. Both solutions used Gaussian mixture models for classification. To evaluate the effectiveness of the proposed solution and compare it with other methods, a voice database was created. The selected EER and minDCF metrics were applied. The experimental results demonstrated the advantages of the proposed algorithm in comparison with the other algorithms. An advantage of the proposed solution is that it uses extracted speech features that perform efficiently when it comes to user identification. This makes it possible to use the algorithm to optimize a voice biometric system that has embedded protection against spoofing attacks that is built on speech synthesis. In addition, it is possible to use the proposed method for voice identification with minimal modifications required. Voice biometric identification systems have excellent opportunities in the banking sector. Such systems allow banks to simplify and accelerate the process of financial transactions and provide their users with advanced banking functions remotely. The implementation of voice biometric systems is difficult by their vulnerability to spoofing attacks, particularly to those conducted by means of speech synthesis. The proposed solution can be integrated into voice biometric systems to improve their security.

Drone classification using RF signal based spectral features

Drone detection and classification, important in military and civilian applications, are performed using different sensor signals. Proposed study handles this task using Radio Frequency (RF) signals utilizing basic machine learning methods. It is composed of two main stages as feature extraction succeeded by training/testing of the model. In feature extraction stage, valuable information for classification, contained in the RF signal, is obtained. For this purpose, spectral features, frequently used in speech processing applications, are employed. Specifically, Power Spectral Density (PSD), Mel-Frequency Cepstral Coefficients (MFCC) and Linear Frequency Cepstral Coefficients (LFCC) are adopted by adjusting filter bank margins and parameters for this task. In the second stage, a Support Vector Machine (SVM) classifier is first trained based on the obtained features and finally tested to measure its performance. All experimental studies are carried out using publicly available DroneRF dataset. This dataset contains 2-Class, 4-Class and 10-Class samples for drone existence vs. background (BG), drone types and drone operation modes, respectively. The best classification results are obtained using, PSD, MFCC and LFCC based features for 2-Class, MFCC and LFCC based features for 4-Class and LFCC based features for 10-Class. Accuracy rates for 2-Class, 4-Class and 10-Class are 100%, 98.67% and 95.15%, respectively. These results show that the proposed method outperforms the results given in the literature for DroneRF dataset.

Hierarchical fusion of visual and physiological signals for emotion recognition

Emotion recognition is an attractive and essential topic in image and signal processing. In this paper, we propose a multi-level fusion method to combine visual information and physiological signals for emotion recognition. For visual information, we propose a serial fusion of two-stage features to enhance the representation of facial expression in a video sequence. We propose to integrate the Neural Aggregation Network with Convolutional Neural Network feature map to reinforce the vital emotional frames. For physiological signals, we propose a parallel fusion scheme to widen the band of the annotation of the electroencephalogram signals. We extract the frequency feature with the Linear-Frequency Cepstral Coefficients and enhance it with the signal complexity denoted by Sample Entropy (SampEn). In the classification stage, we realize both feature level and decision level fusion of both visual and physiological information. Experimental results validate the effectiveness of the proposed multi-level multi-modal feature representation method.

Speaker Recognition Systems in the Last Decade – A Survey

Speaker Recognition Defined by the process of recognizing a person by his\her voice through specific features that extract from his\her voice signal. An Automatic Speaker recognition (ASP) is a biometric authentication system. In the last decade, many advances in the speaker recognition field have been attained, along with many techniques in feature extraction and modeling phases. In this paper, we present an overview of the most recent works in ASP technology. The study makes an effort to discuss several modeling ASP techniques like Gaussian Mixture Model GMM, Vector Quantization (VQ), and Clustering Algorithms. Also, several feature extraction techniques like Linear Predictive Coding (LPC) and Mel frequency cepstral coefficients (MFCC) are examined. Finally, as a result of this study, we found MFCC and GMM methods could be considered as the most successful techniques in the field of speaker recognition so far.

Voice conversion spoofing detection by exploring artifacts estimates

Automatic speaker verification or voice biometrics is an approach to verify the person’s claimed identity through his/her voice. Voice biometrics finds its application in mobile banking and forensics. With the increased usage of speaker verification systems, studying the spoofing threats to speaker verification systems and building proper countermeasure is gaining attention. Spoofing is a genuine challenge as it leads to increase in the false alarm rate, i.e. an impostor is incorrectly accepted as genuine speaker. To make voice biometrics viable for practical applications there is a need to detect spoofing attack. Voice conversion spoofing is a technique where the imposter speaker’s speech is converted to desired speaker’s speech using signal processing approaches. Studies show that voice conversion introduces artifacts in resultant speech, hence, this paper proposes a novel approach to detect voice conversion spoofing attack by estimating artifact estimates from the given speech signal. To obtain artifact estimate from speech signal non-negative matrix factorization based source separation technique is employed. Later, Convolutional Neural Network based binary classifier is built to classify artifact estimates of input speech as natural and synthetic speech. Experiments are conducted on voice conversion challenge 2016 and voice conversion challenge 2018 database. Results show that proposed technique gives excellent performance by detecting wide range of unknown attacks. The proposed systems are compared to state of art spoof detection systems based on Constant Q Cepstrum Coefficients and Linear Frequency Cepstral Coefficients and results show the proposed system give relatively equivalent and/or better performance. Validation results for various noises is studied using NOIZEUS database and results show the efficiency of the proposed system in noisy environments.

Voice Spoofing Countermeasure for Logical Access Attacks Detection

Voice-driven devices (VDDs) like Google Home and Amazon Alexa, which are well-known connected devices in consumer IoT, have applications in various domains i.e., home appliances automation, next-generation vehicles, voice banking, and so on. However, these VDDs that are based on automatic speaker verification systems (ASVs) are vulnerable to voice based logical access (LA) attacks like Text-to-Speech (TTS) synthesis and converted voice signals. Intruders can exploit these attacks to bypass the security of such systems and gain access of victim’s bank account or home control. Thus, there exists a need to develop an effective voice spoofing countermeasure that can reliably be used to protect these VDDs against such malicious attacks. This work presents a novel audio features descriptor named as extended local ternary pattern (ELTP) to capture the vocal tract dynamically induced attributes of bonafide speech and algorithmic artifacts in synthetic and converted speeches. We fused our novel ELTP features with the linear frequency cepstral coefficients (LFCC) to further strengthen the capability of our features for capturing the traits of bonafide and spoofed signals. We employ the proposed ELTP-LFCC features to train the deep bidirectional Long Short-Term Memory (DBiLSTM) network for classification of the bonafide and spoof signal (i.e., TTS synthesis, converted speech). Performance of our spoofing countermeasure is measured on the large-scale and diverse ASVspoof 2019 logical access dataset. Experimental results demonstrate that the proposed audio spoofing countermeasure can reliably be used to detect the LA spoofing attacks.

A Comparison of Cepstral Features in the Detection of Pathological Voices by Varying the Input and Filterbank of the Cepstrum Computation

Automatic voice pathology detection enables objective assessment of pathologies that affect the voice production mechanism. Detection systems have been developed using the traditional pipeline approach (consisting of the feature extraction part and the detection part) and using the modern deep learning -based end-to-end approach. Due to the lack of vast amounts of training data in the study area of pathological voice, the former approach is still a valid choice. In the existing detection systems based on the traditional pipeline approach, the mel-frequency cepstral coefficient (MFCC) features can be regarded as the defacto standard feature set. In this study, automatic voice pathology detection is investigated by comparing the performance of various MFCC variants derived by considering two factors: the input and the filterbank in the cepstrum computation. For the first factor, three inputs (the voice signal, the glottal source and the vocal tract) are compared. The glottal source and the vocal tract are estimated using the quasi-closed phase glottal inverse filtering method. For the second factor, the mel-frequency and linear-frequency filterbanks are compared. Experiments were conducted separately using six databases consisting of voices produced by speakers suffering from one of four disorders (dysphonia, Parkinson’s disease, laryngitis, or heart failure) and by healthy speakers. Support vector machine (SVM) was used as the classifier. The results show that by combining mel- and linear-frequency cepstral coefficients derived from the glottal source and vocal tract, better overall detection accuracy was obtained compared to the defacto MFCC features derived from the voice signal. Furthermore, this combination provided comparable or better performance than four existing cepstral feature extraction techniques in clean and high signal-to-noise ratio (SNR) conditions.

On the use of blind channel response estimation and a residual neural network to detect physical access attacks to speaker verification systems

Spoofing attacks have been acknowledged as a serious threat to automatic speaker verification (ASV) systems. In this paper, we are specifically concerned with replay attack scenarios. As a countermeasure to the problem, we propose a front-end based on the blind estimation of the channel response magnitude and as a back-end a residual neural network. Our hypothesis is that the magnitude response of the channel, obtained by subtracting the log-magnitude spectrum of the observed signal from the prediction of the log-magnitude spectrum average of the observed signal’s clean counterpart, will capture the nuances of room ambiences, recordings and playback devices. The performance of these features is investigated on a benchmark back-end, based on a Gaussian mixture model and on a deep neural network classifier. Our experiments are performed on the 2017 and 2019 Automatic Speaker Verification Spoofing and Countermeasures Challenge (ASVspoof) datasets. The benchmark systems are the same as used in the challenges and are based on constant-Q cepstral coefficients (CQCC) and linear-frequency cepstral coefficients (LFCC) features. Experimental results on the 2017 dataset show that the proposed method outperforms the two benchmarks, providing equal-error rates (EER) as low as 7.57% and 11.64%, respectively, for the development and evaluation sets. On the ASVspoof 2019 dataset, in turn, the proposed method outperformed the benchmark using a residual neural network as back-end by yielding tandem detection cost function (t-DCF) and EER as low as 0.1086 and 4.26% on the evaluation set. Lastly, an instrumental (objective) quality assessment is performed on the two datasets and the impact of quality variability on spoofing detection accuracy is discussed.

Detection of replay spoof speech using teager energy feature cues

The vulnerability of Automatic Speaker Verification (ASV) systems to spoofing or presentation attacks is still an open security issue. In this context, replay spoofing attacks pose a great threat to an ASV system since they can be easily performed (using a playback device, and without needing any technical skill). In this paper, we analyze replay speech signals in terms of reverberation that may occur during recording of the speech signal. Such reverberation introduces delay and changes in amplitude, producing close copies of speech signals, which significantly influences the replay components. To that effect, we propose to exploit the capabilities of the Teager Energy Operator (TEO) to compute a running estimate of subband energies for replay vs. genuine signals. We have used a linearly-spaced Gabor filterbank to obtain a narrowband filtered signal. The TEO has the ability to track the instantaneous changes of a signal. Experiments are performed on the ASVspoof 2017 Challenge version 2.0 database using a Gaussian Mixture Model (GMM) as pattern classifier. Furthermore, we compared our results with state-of-the-art feature sets, namely, Constant Q Cepstral Coefficients (CQCC), Linear Frequency Cepstral Coefficients (LFCC), Mel Frequency Cepstral Coefficients (MFCC), and used their score-level fusion with the proposed feature sets, i.e., Teager Energy Cepstral Coefficients (TECC), in order to obtain possible complementary information that further reduces the Equal Error Rate (EER). Relatively low EERs are obtained with score-level fusion of CQCC, MFCC, LFCC, and TECC feature sets, resulted in 6.68% and 10.45% on development and evaluation sets, respectively. Moreover, for the evaluation dataset, we also studied the performance of the TECC feature set on different Replay Configurations (RC), namely, for acoustic environments, playback, and recording devices. For all the levels of threat conditions (i.e., low, medium, and high-level) to an ASV system, the proposed feature set performed better compared to existing state-of-the-art feature sets. In addition to the ASVspoof 2017 Challenge database, we also performed experiments on other spoofing databases, namely, the ASVspoof 2015 Challenge, BTAS 2016, and ASVspoof 2019 Challenge databases. For all the spoofing databases used in this study, the proposed TECC feature set perform significantly better than the other feature sets.