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.

Two-Way Sign Language Translator for Visually Impaired People

Hearing disabilities affect people all over the world. Nearly 360 million people have a disabling hearing loss. In the recent years, there has been a rapid increase in the number of deaf and dumb victims due to birth defects, accidents and oral diseases. Since deaf and dumb people cannot communicate with a hearing person, they have to depend on some sort of visual communication. Sign Language is not understood by the majority of hearing people. The objective of our project is to form a bridge between the hearing-impaired community and the hearing community; to further commence a two-way communication. We propose a two-way sign language translator that converts Voice to sign language and vice versa in real time. We process the video frame by frame using the OpenCV Library in Python. Furthermore, we use background subtraction method called Gaussian Mixture-based Background/Foreground Segmentation Algorithm on each frame to subtract the background. Contours of this processed image are then passed into a Deep Neural Net (DNN) to classify the frame to the corresponding written language equivalent words. For conversion from sign language to voice, we use basic Natural Language Processing and gtts to accurately conserve the grammar in the sign language. The enrolment rate and literacy among the hearing-impaired children is far below the average for the population at large. This technology will then help the normal schools better integrate the hearing-impaired community thus making education more accessible and cheaper for them. Speech Emotion Recognition (SER) is a vital aspect of human-computer interaction and emotional intelligence applications. This SER that employs Mel- frequency cepstral coefficients (MFCC) as the basis for feature extraction, followed by a Convolutional Neural Network (CNN) with a customized architecture for emotion classification. The methodology 2 involves transforming audio samples into MFCC images, enabling the application of CNNs typically used in computer vision tasks. This fusion of traditional audio analysis with deep learning techniques harnesses the power of MFCC's spectral representation and CNN's spatial pattern recognition. By modifying the CNN architecture, the model can better discern emotional cues in the MFCC images, enhancing SER performance. This strategy offers SER a viable path forward, with the ability to identify complex emotional expressions. In order to increase the accuracy of emotion recognition, it tackles the requirement for feature-rich representations and sophisticated modelling techniques. KEYWORDS: OpenCV, DNN, SER, CNN, MFCC

Classification of Assamese Folk Songs’ Melody using Supervised Learning Techniques

Objectives: A melody is made up of several musical notes or pitches that are joined together to form one whole. This experiment aims to develop four models based on the Mel- frequency Cepstral Coefficients (MFCC) to classify the melodies played on harmonium corresponding to five different class of Assamese folk Music. Methods: The melodies of five different categories of Assamese folk songs are selected for classification. With the help of expert musicians, these melodies are played in harmonium and audio samples are recorded in the same acoustic environment. 20 MFCC’s are extracted from each of the samples and classification of the melodies is done using four supervised learning techniques- Decision Tree Classifier, Linear Discriminant Analysis (LDA), Random Forest Classifier, and Support Vector Machine (SVM). Findings: The performance of the fitted models are evaluated using different evaluation techniques and presented. A maximum of 94.17% average accuracy score is achieved under Support Vector Machine. The average accuracy scores of Decision Tree Classifier, Linear Discriminant Analysis (LDA), and Random Forest Classifier are 73.58%, 85.58%, and 86.11% respectively. The models are developed based on 250 samples (50 from each type). However, increasing the training sample size, there is a possibility to improve the performances of the other three models also. Novelty: The developed approach for identifying the melodies is based on computational techniques. This work will certainly provide a basis for conducting further computational studies in folk music for any community. Keywords: Assamese Folk Music; Decision Tree Classifier; Linear Discriminant Analysis; Random Forest Classifier; Support Vector Machine

Diagnosis of Peripheral Artery Disease Using Backflow Abnormalities in Proximal Recordings of Accelerometer Contact Microphone (ACM).

The development of an accurate, non-invasive method for the diagnosis of peripheral artery disease (PAD) from accelerometer contact microphone (ACM) recordings of the cardiac system. Mel frequency cepstral coefficients (MFCCs) are initially extracted from ACM recordings. The extracted MFCCs are then used to fine-tune a pre-trained ResNet50 network whose middle layers provide streams of high-level-of-abstraction coefficients (HLACs) which could provide information on blood pressure backflow caused by arterial obstructions in PAD patients. A vision transformer is finally integrated with the feature extraction layer to detect PAD, and stratify the severity level. This architecture is coined multi-stream-powered vision transformer (MSPViT). The performance of MSPViT is evaluated on 74 PAD and 21 healthy subjects. Sensitivity, specificity, F1 score, and area under the curve (AUC) of 99.45%, 98.21%, 99.37%, and 0.99, respectively, are reported for the binary classification which ensures accurate detection of PAD. Furthermore, MSPViT suggests average sensitivity, specificity, F1 score, and AUC of 96.66%, 97.34%, 96.29%, and 0.96, respectively, for the classification of subjects into healthy, mild-PAD, and severe-PAD classes. The silhouette score is calculated to assess the separability of clusters formed for classes in the penultimate layer of MSPViT. An average silhouette score of 0.66 and 0.81 demonstrate excellent cluster separability in PAD detection and severity classification, respectively. The achieved performance suggests that the proximal ACM-driven framework can replace state-of-the-art techniques for PAD detection. This study presents a fundamental step towards prompt and accurate diagnosis of PAD and stratification of its severity level.

Analisis Fitur dan Convolutional Neural Network pada Pengenalan Aksen Ucapan

Each country has its characteristics and culture, one of these characteristics is the accent of speech. By listening to someone's accent, we can identify the country of origin of the speaker. Research on accent recognition includes Automatic Speech Recognition (ASR) Technology which is currently being developed, an example of ASR technology, namely Virtual Assistant, the development of this research can be more intelligent Virtual Assistant because it can provide an accent from a speaker. In this study, the authors tried to classify accents from various countries (5 classes), namely English, Spanish, Mandarin, French and Arabic. The dataset used in this study consists of English 627 audio, Spanish 220 audio, Mandarin 132 audio, French 80 audio, and Arabic 172 audio, where all sentences are the same sentence in English. In this study, the audio features used are Mel - Frequency Cepstral Coefficients (MFCC), Zero Crossing Rate (ZCR), and Energy (in librosa it is called RMS). Audio feature extraction generates an array of each audio, the result of audio feature extraction will be the Convolutional Neural Network (CNN) Method input for classifying the accent. This research resulted in 51.30% accuracy for the MFCC feature, 48.05% for the ZCR feature, and 51.95% for the Energy feature. The Energy feature gets good accuracy, followed by the MFCC and ZCR features.

Investigation of Spoken-Language Detection in Multilingual Environment

Spoken language contains lot of information such as information about the content of a message and information about the speaker of that message. Content is composed of several levels of linguistic information like phonological information, morphological information, syntactic information, and the semantic information. For Present study, Multilingual Speech Processing database of different speakers has been recorded in three Indian languages, i.e., Hindi, Marathi, and Rajasthani. The sentences consist of consonants, i.e., “Cha”, “Sha” and “Jha”. Total numbers of speakers involved are 30 including males and females. The basic features of the speech signal: Pitch and first three Formant F1, F2 and F3 are calculated through PRAAT software whereas cepstral features Mel- Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC) has been extracted from MATLAB software. A model is proposed to identify the speaker by multi language speech signal of a speaker using MFCC, GFCC and combine features as acoustic features. For training and testing, is performed on using neural network function Resilient Back Propagation Algorithm and Radial Basis Functions and results are compared. In this experiment accuracy of spoken language identification is 94.77% using BPA and 96.52% using RBF neural network.

Optimization of data-driven filterbank for automatic speaker verification

Most of the speech processing applications use triangular filters spaced in mel-scale for feature extraction. In this paper, we propose a new data-driven filter design method which optimizes filter parameters from a given speech data. First, we introduce a frame-selection based approach for developing speech-signal-based frequency warping scale. Then, we propose a new method for computing the filter frequency responses by using principal component analysis (PCA). The main advantage of the proposed method over the recently introduced deep learning based methods is that it requires very limited amount of unlabeled speech-data. We demonstrate that the proposed filterbank has more speaker discriminative power than commonly used mel filterbank as well as existing data-driven filterbank. We conduct automatic speaker verification (ASV) experiments with different corpora using various classifier back-ends. We show that the acoustic features created with proposed filterbank are better than existing mel-frequency cepstral coefficients (MFCCs) and speech-signal-based frequency cepstral coefficients (SFCCs) in most cases. In the experiments with VoxCeleb1 and popular i-vector back-end, we observe 9.75% relative improvement in equal error rate (EER) over MFCCs. Similarly, the relative improvement is 4.43% with recently introduced x-vector system. We obtain further improvement using fusion of the proposed method with standard MFCC-based approach.

Automatic Speaker Recognition using MFCC and Artificial Neural Network

Automatic speaker recognition is the process of identification of a person automatically from his/her voices. A robust feature extraction algorithm is required for effective and efficient classification. In this paper, a new method is proposed for identifying the speaker using an artificial neural network. Here mel- frequency cepstral coefficient(MFCC) is used as a feature extraction technique that provides useful features for the recognition process. Using these extracted features value, input samples are then created and finally, classification is performed using Multilayer Perceptron (MLP) which is trained by backpropagation. This proposed method gives an accuracy of 94.44%.

Identifying speaker from disguised speech using aural perception and Mel-frequency cepstral coefficient

Objective: The present study intended to compare the accuracy of speaker identification using aural perception and semiautomatic method (Mel –Frequency Cepstral Coefficient; MFCC), when the speech is in disguise condition by using the handkerchief during recording and to check the percentage of correct identification in the semiautomatic method when the vowel and consonant segments were used for analysis. Methods: Thirty speaker's single sentence speech sample was recorded in undisguised and disguised conditions were randomly paired into the sets of one undisguised followed by five disguised samples for the task of speaker identification. In aural perceptual method the five judges listened to the samples and made a decision on the match. In MFCC method, from /ðə/ segment, ten coefficient values were extracted. The coefficient values were manually averaged and the pair that obtained the lowest value of Euclidean distance was determined to be the sample of the same speaker. The Kappa agreement was used to find the agreement between the two methods in speaker identification and the percentage of correct identification was calculated for the vowel and consonant segment analysis. Results: The results revealed the kappa value to be negative (k < 0) indicating no agreement between the two methods. The percentage of correct identification using aural perception ranged from 56.7% - 80% and for MFCC under whole word, consonant segment and vowel segment analysis were 46.7%, 26.7% and 53.33% respectively. Conclusion: The aural perception method had a greater percentage of correct identification than MFCC though it was not statistically significant for speaker identification from disguised speech.

A Pathological Voices Assessment Using Classification

The diagnosing of pathological voice may be a tedious topic and it receives abundant attention. There are several diseases that adversely have an effect on our human speech (voice). The doctor will use only the equipments for detection of pathological voice. However, it is invasive and needs a skilled analysis of diverse human speech signal parameters. Automatic voice analysis for pathological speech has its own blessings, like 1)its quantitative and non-invasive nature. 2) permitting the identification and observance of vocal system diseases . Within the pathological voice classification techniques gathered by the voice of a patient, the goal is to discriminate whether the given voice is normal or pathological. From the speech Mel- Frequency Cepstral Coefficients (MFCC) has been extracted from the voice information and classified into two categories. However, the accuracy of the earlier classification methodology may need additional improvement. In my project work, Support Vector Machine (SVM) classifier is used for pathological voice classification with non-invasive nature to diagnose and analyze the voice of the patient.

Steganalysis of transcoding steganography

Transcoding steganography (TranSteg) is a fairly new IP telephony steganographic method that functions by compressing overt (voice) data to make space for the steganogram by means of transcoding. It offers high stega- nographic bandwidth, retains good voice quality, and is generally harder to detect than other existing VoIP stegano- graphic methods. In TranSteg, after the steganogram reaches the receiver, the hidden information is extracted, and the speech data is practically restored to what was originally sent. This is a huge advantage compared with other existing VoIP steganographic methods, where the hidden data can be extracted and removed, but the original data cannot be re- stored because it was previously erased due to a hidden data insertion process. In this paper, we address the issue of steganalysis of TranSteg. Various TranSteg scenarios and possibilities of warden(s) localization are analyzed with regards to the TranSteg detection. A novel steganalysis method based on Gaussian mixture models and mel- frequency cepstral coefficients was developed and tested for various overt/covert codec pairs in a single warden sce- nario with double transcoding. The proposed method allowed for efficient detection of some codec pairs (e.g., G.711/G.729), while some others remained more resistant to detection (e.g., iLBC/AMR).

Acoustic Signal based Traffic Density State Estimation using SVM

Based on the information present in cumulative acoustic signal acquired from a roadside- installed single microphone, this paper considers the problem of vehicular traffic density state estimation. The occurrence and mixture weightings of traffic noise signals (Tyre, Engine, Air Turbulence, Exhaust, and Honks etc) are determined by the prevalent traffic density conditions on the road segment. In this work, we extract the short-term spectral envelope features of the cumulative acoustic signals using MFCC (Mel- Frequency Cepstral Coefficients). Support Vector Machines (SVM) is used as classifier is used to model the traffic density state as Low (40 Km/h and above), Medium (20-40 Km/h), and Heavy (0-20 Km/h). For the developing geographies where the traffic is non-lane driven and chaotic, other techniques (magnetic loop detectors) are inapplicable. SVM classifier with different kernels are used to classify the acoustic signal segments spanning duration of 20-40 s, which results in average classification accuracy of 96.67% for Quadratic kernel function and 98.33% for polynomial kernel function, when entire frames are considered for classification.

Segmentation of specific speech signals from multi-dialog environment using SVM and wavelet

In this paper, a novel multi-speaker segmentation technique is presented. This technique makes use of wavelets and support vector machines (SVMs) to segment specific speakers’ speech signals from multi-dialog environments. The proposed method first applies wavelets to determine the acoustical features such as subband power and pitch information from a given multi-dialog speech data. Then the multi-speaker segmentation of the given multi-dialog speech data can be accomplished by the use of a bottom–up SVM over these acoustical features and additional parameters, such as frequency cepstral coefficients. A public audio database, Aurora-2, is used to evaluate the performances of the proposed method. Experimental results show that the accuracy of multi-speaker segmentation is 100% achieved in the combination of two speakers. And the segmental accuracy can achieve at least 94.12% and 85.93% for 4-speaker and 8-speaker conditions, respectively.

Audio classification and categorization based on wavelets and support vector Machine

In this paper, an improved audio classification and categorization technique is presented. This technique makes use of wavelets and support vector machines (SVMs) to accurately classify and categorize audio data. When a query audio is given, wavelets are first applied to extract acoustical features such as subband power and pitch information. Then, the proposed method uses a bottom-up SVM over these acoustical features and additional parameters, such as frequency cepstral coefficients, to accomplish audio classification and categorization. A public audio database (Muscle Fish), which consists of 410 sounds in 16 classes, is used to evaluate the performances of the proposed method against other similar schemes. Experimental results show that the classification errors are reduced from 16 (8.1%) to six (3.0%), and the categorization accuracy of a given audio sound can achieve 100% in the Top 2 matches.