Mel-cepstral Coefficients Research Articles

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.

Acousto-optic Multi-source Information Fusion Algorithm for Outdoor Pedestrian Detection

The inefficiency of traditional rescue methods is a serious threat to the life and health of outdoor people in distress, so the research on pedestrian identification for outdoor rescue has important theoretical significance and application value. In order to solve the problem of low probability of recognition based on signal information in traditional methods, an outdoor pedestrian recognition model based on acousto-optic information fusion is constructed. Considering the characteristics of the information source, the directional gradient histogram of the infrared image and the Mel Cepstral Coefficients of the acoustic signal are selected as the factors affecting the recognition probability, and the pedestrian recognition probability correlation equation is deduced; Based on the improved D-S evidence theory of fuzzy system, a pedestrian recognition method based on infrared image and sound signal is proposed, and the mapping relationship between two kinds of information sources and detection probability is established to improve the pedestrian recognition probability. The hardware-in-the-loop simulation results show that the pedestrian recognition probability of fusion of infrared image and sound signal is higher than that of single information, which provides an efficient and reliable method for outdoor rescue.

Research and DSP Implementation of Speech Enhancement Technology Based on Dynamic Mixed Features and Adaptive Mask

A deep learning speech enhancement algorithm based on dynamic hybrid feature and adaptive mask and DSP implementation is proposed in this paper, which solves the problem of feature loss and improves the performance of speech enhancement. The dynamic features incorporate the log Mel power spectrum, Mel cepstral coefficients, and Multiresolution Auditory Cepstral Coefficients (MRACC) and capture the speech transient information by deriving the derivatives to comprehensively represent the nonlinear structure of speech and reduce distortion. To make the system improve the speech quality while reducing the speech distortion as much as possible, a soft mask that can be adaptively adjusted considering the signal-to-noise ratio information is proposed, which can be automatically adjusted according to the different speech signal-to-noise ratio information to obtain the mask value under the corresponding signal-to-noise ratio conditions, and phase difference information that can improve the speech intelligibility is incorporated in it. Then, an improved deep neural network model is designed to effectively improve the speech enhancement performance. Finally, the hardware and algorithm software design of the DSP-based speech enhancement system is given. Experimental simulations are carried out for multiple voices in different noise backgrounds. The experimental results indicate that the performance indexes of the proposed method are significantly improved compared with the existing speech enhancement methods, which verifies the feasibility and superiority of the proposed method.

Comparative Analysis of Machine Learning Algorithms for Classification of Environmental Sounds and Fall Detection

In recent years, number of elderly people in population has been increased because of the rapid advancements in the medical field, which make it necessary to take care of old people. Accidental fall incidents are life-threatening and can lead to the death of a person if first aid is not given to the injured person. Immediate response and medical assistance are necessary in case of accidental fall incidents to elderly people. The research community explored various fall detection systems to early detect fall incidents, however, still there exist numerous limitations of the systems such as using expensive sensors, wearable sensors that are hard to wear all the time, camera violates the privacy of person, and computational complexity. In order to address the above-mentioned limitations of the existing systems, we proposed a novel set of integrated features that consist of melcepstral coefficients, gammatone cepstral coefficients, and spectral skewness. We employed a decision tree for the classification performance of both binary problems and multi-class problems. We obtained an accuracy of 91.39%, precision of 96.19%, recall of 91.81%, and F1-score of 93.95%. Moreover, we compared our method with existing state-of-the-art methods and the results of our method are higher than other methods. Experimental results demonstrate that our method is reliable for use in medical centers, nursing houses, old houses, and health care provisions.

Voice privacy using CycleGAN and time-scale modification

Extensive use of Intelligent Personal Assistants (IPA) and biometrics in our day-to-day life asks for privacy preservation while dealing with personal data. To that effect, efforts have been made to preserve the personally identifiable characteristics from human voice using different speaker anonymization techniques. In this paper, we propose Cycle Consistent Generative Adversarial Network (CycleGAN) to modify (transform) the speaker’s gender as well as the other prosodic aspects using their Mel cepstral coefficients (MCEPs) and fundamental frequency (i.e., F0). For effective anonymization in the context of voice privacy, we propose two-level (i.e., double) anonymization, where first-level anonymization is done using CycleGAN, followed by second-level anonymization using time-scale modification. The speaker anonymization and intelligibility are measured objectively using the automatic speaker verification (ASV) and automatic speech recognition (ASR) experiments, respectively, on development and test sets of Librispeech and VCTK datasets. For CycleGAN-based anonymization, the average % EERs (% WERs) are 40.3% (8.89%) and 40.95% (9.37%) with original enrollments and anonymized trials of the development and test datasets, respectively. The average % EERs (% WERs) for double anonymization are 46.19% (9.95%) and 44.76% (10.34%) with original enrollments and anonymized trials of the development and test datasets, respectively. For the voice privacy evaluation , the performance of ASV system is much important, when the enrollments and trials both are anonymized (called as A-A case), which is also briefly discussed in this work. The average % EERs for A-A case (test set) are 24.29% and 2.81% using CycleGAN-based anonymization and double anonymization, respectively. Objective evaluation for more advanced attack model (i.e., attacker having anonymized data) is also explored in this study. The performance reflected the robustness of proposed anonymization approach towards voice privacy. The subjective tests using 101 listeners and corresponding analysis of variance (ANOVA) and Tukey–Kramer-based Ad-hoc tests are also carried out in order to quote statistical significance of our results. The subjective test show that the CycleGAN and double anonymization approaches give better naturalness, intelligibility, and speaker dissimilarity than the state-of-the-art x-vector-based baseline system.

Improved Power Normalized Cepstrum Coefficient Based on Wavelet Packet Decomposition for Trunk Borer Detection in Harsh Acoustic Environment

The sound-detection method of trunk borer is a very promising method in the field of forestry prevention and control of trunk borers. However, the detection accuracy of commonly used algorithms often decreases sharply in the case of noise reverberation interference. In practical applications, the sound monitoring of trunk borers often takes place in a harsh acoustic environment. To solve this problem, we intend to introduce methods which are effective in other related acoustic fields. Unfortunately, most of the methods are not suitable for acoustic detection of trunk borers and perform extremely poorly. After trying various methods, we found that Power-Normalized Cepstral Coefficients (PNCC) performed well in some cases, while it did not in others. This is due to the difference between speech and trunk borer sound. Therefore, an improved anti-noise PNCC based on wavelet package is proposed. The dmey wavlet system always obtains the best performance. We collected the audio of the following five dry borer pests for testing. They are red palm weevil, mountain pine beetle, red necked longicorn, Asian longhorn beetle and citrus longhorn beetle. In the experimental part, we used genetic algorithm-support vector machine (GA-SVM) as a classifier to compare Mel Cepstral Coefficients (MFCC), which are the most common methods in the field of audio detection of trunk borer, PNCC and improved PNCC in a variety of noise environments. The results showed that, compared with other methods, the newly proposed method can often achieve better results. The above experiments take the audio clips made of clear pest sound mixed noise. In order to further verify the effectiveness of the method, we designed another experiment with a harsh outdoor acoustic environment. We found that the proposed method achieved 88% accuracy and the traditional PNCC achieved 78% accuracy. However, the Mel cepstrum coefficient completely lost its ability to distinguish. In sum, the proposed PNCC based on wavelet packet decomposition can be used as a detection method for trunk borer in the harsh acoustic environment. This method has many advantages, including simple extraction and strong robustness to noise. Combined with cheap audio acquisition equipment, this method can effectively improve the early warning ability of forestry borer pests.

Convolution neural network based automatic speech emotion recognition using Mel-frequency Cepstrum coefficients

A significant role is played by Speech Emotion Recognition (SER) with different applications in affective computing and human-computer interface. In literature, the most adapted technique for recognition of emotion was based on simple feature extraction using a simple classifier. Most of the methods in the literature has limited efficiency for the recognition of emotion. Hence for solving these drawbacks, five various models based on Convolution Neural Network (CNN) was proposed in this paper for recognition of emotion through signals obtained on speech. In the methodology which was proposed, seven different emotions are recognised with the utilisation of CNN with feature extraction methods includes disgust, normal, fear Joy, Anger, Sadness and surprise. Initially, the speech emotion signals are collected from the database such as berlin database. After that, feature extraction is considered, and it is carried out by the Pitch and Energy, Mel-Frequency Cepstral Coefficients (MFCC) and Mel Energy Spectrum Dynamic Coefficients (MEDC). The mentioned feature extraction process is widely used for classifying the speech data and perform better in performance. Mel-cepstral coefficients utilise less time for shaping the spectral with adequate data and offers better voice quality. The extracted features are used for the recognition purpose by CNN network. In the proposed CNN network, either one or more pairs of convolutions, besides, max-pooling layers remain present. With the utilisation of the CNN network, the emotions are recognised through the input speech signal. The proposed method is implemented in MATLAB, and it will be contrasted with the existing method such as Linear Prediction Cepstral Coefficient (LPCC) with the K-Nearest Neighbour (KNN) classifier to test the samples for optimal performance evaluation. The Statistical measurements are utilised for analysing the performance such as accuracy, precision, specificity, recall, sensitivity, error rate, receiver operating characteristics (ROC) curve, an area under curve (AUC), and False Positive Rate (FPR).

Acoustic features of vocalization signal in poultry health monitoring

In precision livestock farming, studies show that bird sound can be employed as a biomarker of health condition. One of the most important steps for this purpose is to study the feasibility of using acoustic features as criteria for disease diagnosis. In this research five acoustic features of bird calls were evaluated for determination of bird health condition. Signals were collected from broilers grown in three groups: control, challenged with Bronchitis, and challenged with Newcastle disease. Results of data analysis showed that, among the 5 acoustic features studied, wavelet entropy (WET) had the best performance and was able to detect Bronchitis on the third day after inoculation with 83% accuracy while the type II error in this test (incorrectly detecting sick bird as healthy) was less than 14% and 6% on the third day and fourth day, respectively. In the case of Newcastle disease, although WET and Mel cepstral coefficients (MFCC) exhibited similar accuracy (80% and 78% respectively on the fourth day), but the difference was that WET was more reliable in detecting healthy birds while MFCC had better performance detecting challenged birds.

Automatic music transcription based on convolutional neural network, constant Q transform and MFCC

Automatic music transcription is a function that relies on the efficiency of the computer’s own calculation speed to recognize the pitch of external input audio and output it accurately. In the past methods, there are often problems such as inability to reflect music continuity or narrow identifiable frequency range. In order to avoid such problems, two methods different from the previous Mel Cepstral Coefficients and Constant Q Transform are selected to extract the features of music, and convolutional neural networks are used for training and recognition. Among them, the Mel Cepstrum coefficient is used to judge the timbre, and the constant Q transform is used to judge the pitch. After inputting the corresponding features into the neural network for training and learning, we can find that the recognition success rate has reached 95%.

Modelowanie zjawisk wibroakustycznych z zastosowaniem metody parametryzacji sygnału fonicznego

The article proposes an original way of modeling vibroacoustic phenomena of exploited machines/devices using the method of audio parameterization. This method extends the current approach to this type of research and consists in taking into account the psychoacoustic effects associated with the emission of vibroacoustic energy. The proposed solution is based on the determination of mel-cepstral coefficients of the examined signal and its classification, due to the impact of noise. It was presented verification of the method on the example of studies on the impact of road noise sources.

NEURAL NETWORK VOICE RECOGNITION MODEL

The article is devoted to the development of recognition tools for the emotional state of the speaker. The prospects of using neural networks for the analysis of fixed fragments of a voice signal is shown. The necessity of adapting the appearance and parameters of the neural network model to the conditions of the task of recognizing emotions by voice is established. As a result of the studies, it was determined that in the task of recognizing the speaker’s emotions by voice fragments of a fixed duration, it is advisable to use a two-layer perceptron, the input parameters of which are associated with mel-cepstral coefficients characterizing each of the quasi-stationary fragments of the analysed voice signal, and the output parameters correspond to the recognizable emotions of the speaker. The feasibility of using a two-layer perceptron is confirmed by computer experiments. It was determined that the directions of further research are related to determining the number of mel-cepstral coefficients, which is sufficient to describe a single quasistationary fragment, and adapting the parameters of the two-layer perceptron to recognition conditions under the influence of various kinds of interference.

Analysis of vowel production in Mandarin/Hindi/American- accented English for accent recognition systems

With globalization and emerging voice-activated systems, accent recognition systems have gained more importance. Foreign-accented English shows different acoustical characteristics from native English pronunciation. It varies based on native language of speakers. This work investigated the similarities and differences between spectral and time-domain characteristics of vowel production for English /hvd/ words spoken by native Mandarin, Hindi, and American English speakers. Fundamental frequency, the first four formants, mel-cepstral coefficients, linear predictive coefficients, harmonicity, spectral centroid, spectral spread, tonality, spectral flatness, pitch range coefficients, and zero crossings rate were examined for male and female speakers of English. One-way ANOVA was performed to find the significant correlations. Lindblom’s perceptual distance of the corner vowels was calculated for further analysis. Our preliminary findings indicated that the vowel spaces of the male L2 speakers were smaller than the male L1 speakers’ vowel space. The F1-F2 space of the corner vowel /ɑ/ presented the most variance among L1 and L2 speakers. The vowel space of the female Hindi-accented English was found to be nearly triangular shaped. For the vowels /i/ and /ɪ/, Mandarin- and Hindi-accented English showed similar characteristics. Short-time energy had significant means between L1 and L2 speakers (p < 0.001) for the vowels /i/ as in ‘heed’ and /ɪ/ as in ‘hid’. Higher F1 frequencies were calculated for the vowels /ӕ/ as in ‘had’, /ε/ as in ‘head’ and /ʌ/ as in ‘hud’ spoken by Mandarin-accented speakers. Time-domain based features provided noteworthy differences between the female L1 and L2 speakers of English. The baseline classification system showed that spectral features had higher impacts on recognizing Mandarin- and Hindi-accented English.

Optimisation of training samples in recognition of overlapping speech and identification of speaker in a two speakers situation

Recognition of overlapping speech is still a challenging problem in the area of automatic speech recognition (ASR). In this paper, we have proposed a technique for overlapping speech recognition integrated with firefly optimization technique. Overlapped segments are thoroughly analysed for different dominant frequencies involved in the mixture. We have created an audio splitting function. Split audio signals are converted into mel cepstral coefficients and the intensity variations of signals are indicated by their cepstrum. Phoneme density updated cepstrum (PDUC) features are extracted from both spectrum feature analysis and mel frequency cepstral coefficients (MFCC) coefficients. Further, firefly optimization technique is used for clustering and selecting best relevant features. Datasets of speech separation challenge (SSC) Scopus are used to evaluate the results. From the results, we could conclude that minimum samples of 20% to 30% are sufficient to achieve recognition accuracy of above 90%.

Hierarchical Expert Neural Network System for Speech Recognition

This work proposes a hierarchical architecture composed of a expert neural network set based on the ensemble method with dynamic selection of classifiers for application in speech recognition systems. Therefore, 30 commands in the Brazilian Portuguese language were coded by a two-dimensional time matrix, resulting from the application of the discrete cosine transformation in the mel-cepstral coefficients. These patterns were modified by means of a nonlinear transformation to a high-dimensionality space through a set of Gaussian radial basis functions (GRBFs) parameterized with the centroid and covariance characteristics of the classes. The classification was made through the dynamic classifier selection approach, in which multilayer perceptron and learning vector quantization configurations were analyzed to constitute the multiple classifiers specialized in the subdivisions made in the total of classes to be recognized. Then, given a new test pattern, the GRBF that presents the highest value of the receptive field in relation to the input feature vector indicates the class to which the pattern is nearer, thus directing to the expert neural network that provides the final result of classification based on the local accuracy.

Mel-Cepstrum-Based Quantization Noise Shaping Applied to Neural-Network-Based Speech Waveform Synthesis

This paper presents a mel-cepstrum-based quantization noise shaping method for improving the quality of synthetic speech generated by neural-network-based speech waveform synthesis systems. Since mel-cepstral coefficients closely match the characteristics of human auditory perception, the proposed method effectively masks the white noise introduced by the quantization typically used in neural-network-based speech waveform synthesis systems. The paper also describes a computationally efficient implementation of the proposed method using the structure of the mel-log spectrum approximation filter. Experiments using the WaveNet generative model, which is a state-of-the-art model for neural-network-based speech waveform synthesis, showed that speech quality is significantly improved by the proposed method.

An on-line VAD based on Multi-Normalisation Scoring (MNS) of observation likelihoods

Voice activity detection (VAD) is an essential task in expert systems that rely on oral interfaces. The VAD module detects the presence of human speech and separates speech segments from silences and non-speech noises. The most popular current on-line VAD systems are based on adaptive parameters which seek to cope with varying channel and noise conditions. The main disadvantages of this approach are the need for some initialisation time to properly adjust the parameters to the incoming signal and uncertain performance in the case of poor estimation of the initial parameters. In this paper we propose a novel on-line VAD based only on previous training which does not introduce any delay. The technique is based on a strategy that we have called Multi-Normalisation Scoring (MNS). It consists of obtaining a vector of multiple observation likelihood scores from normalised mel-cepstral coefficients previously computed from different databases. A classifier is then used to label the incoming observation likelihood vector. Encouraging results have been obtained with a Multi-Layer Perceptron (MLP). This technique can generalise for unseen noise levels and types. A validation experiment with two current standard ITU-T VAD algorithms demonstrates the good performance of the method. Indeed, lower classification error rates are obtained for non-speech frames, while results for speech frames are similar.

An Anomalous Noise Events Detector for Dynamic Road Traffic Noise Mapping in Real-Life Urban and Suburban Environments.

One of the main aspects affecting the quality of life of people living in urban and suburban areas is their continued exposure to high Road Traffic Noise (RTN) levels. Until now, noise measurements in cities have been performed by professionals, recording data in certain locations to build a noise map afterwards. However, the deployment of Wireless Acoustic Sensor Networks (WASN) has enabled automatic noise mapping in smart cities. In order to obtain a reliable picture of the RTN levels affecting citizens, Anomalous Noise Events (ANE) unrelated to road traffic should be removed from the noise map computation. To this aim, this paper introduces an Anomalous Noise Event Detector (ANED) designed to differentiate between RTN and ANE in real time within a predefined interval running on the distributed low-cost acoustic sensors of a WASN. The proposed ANED follows a two-class audio event detection and classification approach, instead of multi-class or one-class classification schemes, taking advantage of the collection of representative acoustic data in real-life environments. The experiments conducted within the DYNAMAP project, implemented on ARM-based acoustic sensors, show the feasibility of the proposal both in terms of computational cost and classification performance using standard Mel cepstral coefficients and Gaussian Mixture Models (GMM). The two-class GMM core classifier relatively improves the baseline universal GMM one-class classifier F1 measure by 18.7% and 31.8% for suburban and urban environments, respectively, within the 1-s integration interval. Nevertheless, according to the results, the classification performance of the current ANED implementation still has room for improvement.

Improved voicing decision using glottal activity features for statistical parametric speech synthesis

A method to improve voicing decision using glottal activity features proposed for statistical parametric speech synthesis. In existing methods, voicing decision relies mostly on fundamental frequency F0, which may result in errors when the prediction is inaccurate. Even though F0 is a glottal activity feature, other features that characterize this activity may help in improving the voicing decision. The glottal activity features used in this work are the strength of excitation (SoE), normalized autocorrelation peak strength (NAPS), and higher-order statistics (HOS). These features obtained from approximated source signals like zero-frequency filtered signal and integrated linear prediction residual. To improve voicing decision and to avoid heuristic threshold for classification, glottal activity features are trained using different statistical learning methods such as the k-nearest neighbor, support vector machine (SVM), and deep belief network. The voicing decision works best with SVM classifier, and its effectiveness is tested using the statistical parametric speech synthesis. The glottal activity features SoE, NAPS, and HOS modeled along with F0 and Mel-cepstral coefficients in Hidden Markov model and deep neural network to get the voicing decision. The objective and subjective evaluations demonstrate that the proposed method improves the naturalness of synthetic speech.

Emotional voice conversion using neural networks with arbitrary scales F0 based on wavelet transform

An artificial neural network is an important model for training features of voice conversion (VC) tasks. Typically, neural networks (NNs) are very effective in processing nonlinear features, such as Mel Cepstral Coefficients (MCC), which represent the spectrum features. However, a simple representation of fundamental frequency (F0) is not enough for NNs to deal with emotional voice VC. This is because the time sequence of F0 for an emotional voice changes drastically. Therefore, in our previous method, we used the continuous wavelet transform (CWT) to decompose F0 into 30 discrete scales, each separated by one third of an octave, which can be trained by NNs for prosody modeling in emotional VC. In this study, we propose the arbitrary scales CWT (AS-CWT) method to systematically capture F0 features of different temporal scales, which can represent different prosodic levels ranging from micro-prosody to sentence levels. Meanwhile, the proposed method uses deep belief networks (DBNs) to pre-train the NNs that then convert spectral features. By utilizing these approaches, the proposed method can change the spectrum and the F0 for an emotional voice simultaneously as well as outperform other state-of-the-art methods in terms of emotional VC.

Identification Features Analysis in Speech Data Using Gmm-Ubm Speaker Verification System

This paper is devoted to feature selection and evaluation in an automatic text-independent speaker verification task. In order to solve this problem a speaker verification system based on the Gaussian mixture model and the universal background model (GMM-UBM system) was used. The application sphere and challenges of modern systems of automatic speaker identification were considered. Overview of the modern speaker recognition methods and main speech features used in speaker identification is provided. Features extraction process used in this article was examined. Reviewed speech features were used for speaker verification including mel-cepstral coefficients (MFCC), line spectral pairs (LSP), perceptual linear prediction cepstral coefficients (PLP), short-term energy, formant frequencies, fundamental frequency, voicing probability, zero crossing rate (ZCR), jitter and shimmer. The experimental evaluation of the GMM-UBM system using different speech features was conducted on a 50 speaker set and a result is presented. Feature selection was done using the genetic algorithm and the greedy adding and deleting algorithm. Equal error rate (EER) equals 0,579 % when using 256 component Gaussian mixture model and the obtained feature vector. Comparing to standard 14 MFCC vector, 42,1 % of EER improvement was acquired.