Emerging features for speaker recognition

Currently, empirical mode decomposition (EMD) has become a popular data-driven time-frequency analysis method for nonstationary and nonlinear data. However, it is still limited to univariate data due to the number and/or scale misalignment for multivariate data. A newly developed multivariate EMD (MEMD) scheme decomposes multivariate data simultaneously and thus leads to mode alignment and minimizes mode mixing. In addition, an improved amplitude and frequency modulation (AM-FM) decomposition algorithm presented here provides an estimation of a more meaningful instantaneous amplitude and frequency than the widely used Hilbert transform (HT). Both of these facilitate development of a time-frequency analysis framework for multivariate nonstationary and nonlinear data analysis. In this paper, MEMD-based scalogram and coscalogram, and instantaneous frequency spectra and cospectra are proposed to characterize a multivariate nonstationary process. The scalogram and instantaneous frequency spectra capture spectral evolution of each component while the coscalogram and instantaneous frequency cospectra reveal embedded intermittent correlation between two components. Compared with scale-based scalogram and coscalogram, frequency-based instantaneous frequency spectra and cospectra provide a more detailed portrait of multivariate data. The effectiveness of the proposed MEMD-based time-frequency analysis framework is demonstrated by numerical examples of a thunderstorm downburst and an earthquake ground motion. Also, the results from the MEMD-based approach are compared with those based on a continuous wavelet transform, which further reinforces the effectiveness of the proposed framework.

Planetary gearbox vibration signals have strong modulation features due to the amplitude modulation and frequency modulation (AM-FM) effect of gear faults, as well as the amplitude modulation (AM) effect of time-varying vibration transfer paths, on gear meshing vibrations. This results in an involute sidebands structure in Fourier spectrum, possibly misleading fault diagnosis. The modulating frequency of both amplitude modulation (AM) and frequency modulation (FM) parts is closely related to the gear fault characteristic frequency. This inspires the idea of joint amplitude and frequency demodulation analysis, thus addressing the complex sidebands issue inherent in Fourier spectrum. Demodulation analysis requires mono-component signals for accurate estimation of instantaneous frequency, and proper selection of an AM-FM component sensitive to gear fault. To this end, we firstly decompose the complex signal into intrinsic mode functions (IMFs) via variational mode decomposition (VMD), by exploiting its capability in decomposing complex modulated signal into constituent AM-FM components. For effective application of VMD in complex planetary gearbox signal analysis, we propose a method to determine a key parameter in VMD, i.e. the number of IMFs to be separated. For accurate instantaneous frequency estimation, we decompose IMFs via empirical AM-FM decomposition, to remove the influence of AM on instantaneous frequency estimation. Then, we select the sensitive IMF that contains the main gear fault information for further demodulation analysis. In order to properly select the sensitive IMF, we propose a criterion based on the gear vibration characteristics and the VMD properties. Finally, we obtain the amplitude and frequency demodulated spectra by applying Fourier transform to the amplitude envelope and instantaneous frequency of the selected sensitive IMF. According to the characteristics exhibited in the demodulated spectra, we can detect planetary gearbox fault. The proposed method is illustrated via a numerical simulated planetary gearbox vibration signal, and is further validated using lab experimental vibration signals of a planetary gearbox. Faults on all the three types of gear (sun, planet and ring) are successfully identified.

Significance of analytic phase of speech signals in speaker verification

This work focuses on a signal-based diagnosis approach dedicated to proton exchange membrane water electrolyzer (PEM WE) anode pump fault. The PEM WE cell measurements are performed with an experimental test bench to highlight the impact of water flow rate in the anode compartment. This approach is non-intrusive, and it can detect anode flow rate variation during the electrolysis and is designed to fulfill online diagnosis requirements. Contrary to electrochemical impedance spectroscopy-based approaches (EIS), this method stands out from existing procedures as a result of its few requirements, excluding any signal with perturbing amplitude. Therefore, the electrolyzer remains continuously available, even while the analysis is performed. The empirical mode decomposition (EMD) is used to decompose the signal variation into a sum of amplitude modulation and frequency modulation (AM-FM) components, called intrinsic mode functions (IMFs). In this work, the PEM WE current signal is decomposed into several IMFs using EMD. Then, the energetic contribution of each IMF is calculated. Experimental results exhibited that the energetic contribution of IMFs can be used as relevant criteria for fault diagnosis in PEM WE systems. This process only requires monitoring of the PEM WE current and has a low computational cost, which is a significant economic and technical advantage.

ABSTRACT: Speech is one of the most natural and efficient forms of communication among humans, and it is becoming increasingly integral to human-computer interaction in modern applications such as virtual assistants, smart devices, and automated customer service systems. Among the various technologies that leverage speech, speaker recognition stands out as a biometric method that identifies or verifies individuals based on the unique characteristics of their voice. This approach has gained significant traction due to its non-intrusive nature, ease of integration, and applicability across various domains such as security, forensics, and personalised user experiences. In this paper, we introduce a novel speaker recognition system specifically designed to identify speakers based on utterances in the Marathi language, a linguistically rich and widely spoken regional language in India. The system utilises Mel-Frequency Cepstral Coefficients (MFCCS) to extract distinguishing vocal features that closely mimic the human auditory perception. MFCCS are particularly effective in capturing speech’s phonetic and acoustic properties, making them a preferred choice in speech and speaker recognition tasks. Vector quantisation (VQ) is applied to optimise the feature set and reduce computational complexity. VQ compresses the high-dimensional MFCC feature vectors into representative clusters without significantly compromising accuracy, thereby enhancing system efficiency. These features are then processed using hidden Markov Models (HMMS), which are adept at modelling temporal sequences and dynamic variations in speech. HMMS offer a statistical framework that effectively captures the sequential nature of speech patterns, leading to more reliable speaker modelling and recognition.

Gear vibration signals can be modeled as amplitude modulation and frequency modulation (AMFM) processes, and have cyclostationarity, so cyclic spectral analysis is used to extract the modulation features of gearbox vibration signals, and to detect and assess localized gear damage. The cyclic spectral density of AMFM signals is deduced, and its properties in joint cyclic frequency-frequency domain are summarized. The ratio between the sum of cyclic spectral density magnitude along frequency axis at the cyclic frequencies of modulating frequency and 0 Hz can indicate the amplitude modulation magnitude, so it can be used as an indicator to assess the health status of gearboxes. The analysis of a gearbox run-to-failure test shows that the ratio has a monotonically increasing trend with the development of the gear damage. It has potential to early detect incipient damage, to prognosticate severe damage, and to assess gearbox damage.

Gear vibration signals are mainly due to the gear rotating, gear pair meshing and their coupling effects, and are usually modeled as amplitude modulation and frequency modulation (AMFM) processes. The modulation feature of vibration signals is related to the health status of gears. As well, their statistics change periodically with gear rotation, i.e. they are cyclostationary, so cyclostationary analysis is suitable to extract the modulation features of gear vibration signals, so as to detect and assess gear damage. The cyclic auto-correlation function of AMFM signals is deduced, and its properties in cyclic frequency domain are summarized. The ratio between the cyclic autocorrelation slice magnitude at cyclic frequencies of modulating frequency and 0 Hz is a monotonically increasing function of the modulation magnitude, so it is able to indicate the modulation magnitude. Since the modulation magnitude of gear vibration signals is related to localized gear damage degree, the ratio is used to assess the condition of gears. In the analysis of gear crack simulation vibration signals, it shows a monotonically increasing trend with the development of the gear damage. It has potential to early detect incipient damage, to prognosticate severe damage, and to assess gearbox damage.

In this paper, amplitude and frequency modulation (AM-FM) data are extracted from satellite images and processed to facilitate meteorological observations. The approach used is mainly based on 2D-DESA (Discrete Energy Separation Algorithm) where the Teager Kaiser Energy Operator (TKEO) acts as data discriminator. Dominant energy parameters selected by this operator, are employed to get the AM-FM components and segment the images under study via K-means classifier. The AM-FM method so described has been tested on a set of multispectral satellite images collected by Meteosat 9 (MSG-2) over North Africa in November 2011. Applied to both visible and infrared Meteosat Second Generation (MSG) images, it yields a compact and coherent segmentation map of Northern Algeria, Mediterranean Sea and South Western Europe, where the object contours are satisfactorily reproduced and typical clouds, detected.

A new approach for linear prediction (LP) analysis is explored, where predictor can be computed from a mel-warped subband-based autocorrelation functions obtained from the power spectrum. For spectral representation a set of multi-resolution cepstral features are proposed. The general idea is to divide up the full frequency-band into several subbands, perform the IDFT on the mel power spectrum for each subband, followed by Durbin's algorithm and the standard conversion from LP to cepstral coefficients. This approach can be extended to several levels of different resolutions. Multi-resolution feature vectors, formed by concatenation of the subband cepstral features into an extended feature vector, are shown to yield better performance than the conventional mel-warped LPCCs over the full voice-bandwidth for a connected digit recognition task.

We propose a robust event-based method for estimation of the instantaneous fundamental frequency of a voiced speech signal. The amplitude and frequency modulated (AM-FM) signal model of voiced speech in the low frequency range (LFR) indicates the presence of energy only around its instantaneous fundamental frequency ( F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> ) and its few harmonics. The time-varying F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> component of a voiced speech signal is extracted by a robust algorithm which iteratively performs eigenvalue decomposition (EVD) of the Hankel matrix, initially constructed from samples of the LFR filtered voiced speech signal. The negative cycles of the extracted time-varying F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> component provide a reliable coarse estimate of intervals where glottal closure instants (GCIs) may be present. The negative cycles of the LFR filtered voiced speech signal occurring within these intervals are isolated. There is a sudden decrease in the glottal impedance at GCIs resulting in high signal strength. Therefore, GCIs are detected as local minima in the derivative of the falling edges of the isolated negative cycles of the LFR filtered voiced speech signal, followed by a selection criterion to discard false GCI candidates. The instantaneous F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> is estimated as the inverse of the time interval between two consecutive GCIs. Experiments were performed on the Keele and CSTR speech databases in white and babble noise environments at various levels of degradation to assess the performance of the proposed method. The proposed method substantially reduces the gross F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> estimation errors in comparison to some state of the art methods.

Instantaneous frequency(IF) calculation is a key step for empirical mode decomposition(EMD).But signal analysis accuracy is not very well for practical condition signal analysis based on traditional IF calculation method.Although non-stationary signals can be modeled as polynomial phase signals(PPS),traditional phase model estimation methods are not very suitable for practical application,such as maximum likelihood estimation and the least squares method.In most circumstances,a lot of calculations are needed.Thus,several problems need to be solved during IF calculation for intrinsic mode function obtained from EMD.The boundary effect for signal analysis is very obvious.High-order ambiguity function(HAF) is put forward for IF calculation.Then,Hilbert time-frequency spectrum can be constructed for signal analysis.HAF can reduce computation and improve IF estimation accuracy.Simulation signal and practical monitored signal are used to testify the effectiveness of this method for IF calculation.The research shows that this method can effectively eliminate the influence of boundary effect and improve the reliability of Hilbert time-frequency spectrum.

Speaker recognition is an Artificial Intelligent (AI) technology that lets the machine to process, interpret and respond to human language. In this work, the recorded speech developed from a collection of audio speeches is used as a database. Mel-frequency cepstral coefficients (MFCC) and gammatone frequency cepstral coefficients (GFCC) are two different cepstral features used in this work. These extracted features are then used to train, validate and test the classifier. Support Vector Machine (SVM) is the classifier used in developing the speaker identification system. This classifier is trained to classify the input speech into one of the ethnicity classes: Malay, Chinese, Indian or Bumiputera. The results are based on the two different usages of cepstral features from the same speech utterances by speakers. Finally, the comparative analysis of the speaker identification system is made concerning features and classifier. The results revealed that a combination of GFCC and pitch as the feature vectors (Model 4) produced the highest accuracy rate of 86.1%.

Automatic speaker recognition determines whether a person is who he/she claims to be without the intervention of another human being, from their voiceprint. In recent years different methods of voice analysis to fulfill this objective have been developed. At the moment, the development of computational auditory models that emulate the physiology and function of the inner ear have added a new tool in the field of speaker recognition, including the Triple Resonance Nonlinear filter. This paper studies the behavior of a bio-inspired model of inner ear applied to the speakers's voice analysis. Their ability on speaker recognition tasks was statistically analyzed. We conclude, given the results obtained, that this system is an excellent tool, which shows a high rate of sensitivity and specificity in speaker recognition.

This paper compares a set of large scale support vector machine (SVM) training algorithms for language and speaker recognition tasks. We analyze five approaches for training phonetic and acoustic SVM models for language recognition. We compare the performance of these approaches as a function of the training time required by each of them to reach convergence, and we discuss their scalability towards large corpora. Two of these algorithms can be used in speaker recognition to train a SVM that classifies pairs of utterances as either belonging to the same speaker or to two different speakers. Our results show that the accuracy of these algorithms is asymptotically equivalent, but they have different behavior with respect to the time required to converge. Some of these algorithms not only scale linearly with the training set size, but are also able to give their best results after just a few iterations. State-of-the-art performance has been obtained in the female subset of the NIST 2010 Speaker Recognition Evaluation extended core test using a single SVM system.

Automatic speech and speaker recognition are traditionally treated as two independent tasks and are studied separately. The human brain in contrast deciphers the linguistic content, and the speaker traits from the speech in a collaborative manner. This key observation motivates the work presented in this paper. A collaborative joint training approach based on multitask recurrent neural network models is proposed, where the output of one task is backpropagated to the other tasks. This is a general framework for learning collaborative tasks and fits well with the goal of joint learning of automatic speech and speaker recognition. Through a comprehensive study, it is shown that the multitask recurrent neural net models deliver improved performance on both automatic speech and speaker recognition tasks as compared to single-task systems. The strength of such multitask collaborative learning is analyzed, and the impact of various training configurations is investigated.