MFCC (Mel Frequency Cepstral Coefficient)

Mel Frequency Ceptral Coefficient is a very common and efficient technique for signal processing. This paper presents a new purpose of working with MFCC by using it for Hand gesture recognition. The objective of using MFCC for hand gesture recognition is to explore the utility of the MFCC for image processing. Till now it has been used in speech recognition, for speaker identification. The present system is based on converting the hand gesture into one dimensional (1-D) signal and then extracting first 13 MFCCs from the converted 1-D signal. Classification is performed by using Support Vector Machine. Experimental results represents that proposed application of using MFCC for gesture recognition have very good accuracy and hence can be used for recognition of sign language or for other household application with the combination for other techniques such as Gabor filter, DWT to increase the accuracy rate and to make it more efficient.

One of the crucial aspects of the environmental protection is continuous monitoring of environment. Specific aspect is estimation of the bird species population. It is particularly important for bird species being in danger of extinction. Avian monitoring programs are time and money consuming actions which usually base on terrain expeditions. Certain remedy for this can be automatic acoustical avian monitoring system, described in the paper. Main components of the designed system are: digital audio recorder for bird voices acquisition, computer program automatically recognizing bird species by its signals emitted (voices or others) and object-relational database accessed via the Internet. Optional system components can be: digital camera and camcorder, bird attracting device, wireless data transmission module, power supply with solar panel, portable weather station. The system records bird voices and sends the recordings to the database. Recorded bird voices can be also provoked by ...

Birds are a reflection of environmental health as pollution and climate change affect biodiversity. Experts in ecology and machine learning stand to benefit the most from largescale monitoring of biodiversity. Today, convolutional neural networks (CNNs) are the preferred choice for species recognition as their performance has consistently outperformed humans. However, CNNs are disadvantaged by their high computational complexity and the need to provide vast amounts of training data. This paper compares the performance versus the complexity of two widely used CNNs, namely ResNet-50 and MobileNetV1. ResNet-50 is a high-complexity CNN while MobilenetV1 is a low-complexity CNN targeted for mobile applications. We used spectrogram images of Brazilian bird sounds as inputs to both networks. These birds were chosen due to their abundance of samples in the Xeno-canto bird sound repository. Short-Time Fourier Transform (STFT) and Mel Frequency Cepstral Coefficient (MFCC) algorithms are used to extracting spectrogram images. To validate the precision of the classifier, 1,000 spectrogram images of each of ten bird species are produced and fed into both classifiers. The findings indicate that the accuracy of MobileNetV1 is close to that of ResNet-50, with MFCC which is 85.73 and 90.56 respectively.

Искусственные нейронные сети в MATLAB: учебное пособие / В.К. Тытюк, В.В. Бушер, О.П. Черный; КНУ – Кривой Рог, 2025. – 98 с.

Дается краткое представление о работе в системе MATLAB 2024, краткое теоретическое введение в курс искусственных нейронных сетей (ИНС, ANN). Рассматриваются вопросы подготовки данных для работы с ИНС, также вопросы проектирования структур ИНС, их обучения и использования для задач обработки данных. Пособие формирует навыки работы с известными типами нейронных сетей в системе MATLAB 2024.

Лабораторный практикум предназначен для студентов при изучении дисциплин «Нейрокомпьютерные системы», «Нейросетевые технологии», «Интеллектуальные системы управления».
Также рекомендуется студентам, специализирующимся в области искусственного интеллекта и обработки больших данных (BigData).

Se plantea una solución al problema de reconocer las vocales A-I-U usando un clasificador bayesiano. Este es un problema interesante de abordar debido a que constituye la base para implementar sistemas de transcripción automática de texto. El presente documento se centra en encontrar el subconjunto mínimo de coeficientes MFCC para clasificar las vocales planteadas. Se determinó que sólo tres de los trece coeficientes son necesarios para que un clasificador bayesiano alcance una precisión del 99.2% en las grabaciones empleadas.

Tertawa adalah respons vokal yang umum dalam interaksi sosial, yang dapat mempengaruhi karakteristik suara individu. Penelitian ini bertujuan membandingkan karakteristik sinyal suara sebelum dan sesudah tertawa. Subjek penelitian terdiri dari empat perempuan dewasa berusia 20 tahun, dengan pengukuran yang dilakukan pada beberapa parameter suara, termasuk amplitudo, frekuensi dasar (F0), dan timbre. Data dikumpulkan menggunakan rekaman suara digital sebelum dan sesudah aktivitas tertawa, kemudian dianalisis menggunakan perangkat lunak analisis suara. Hasil penelitian menunjukkan adanya peningkatan rata-rata amplitudo sebesar 0,2 dB setelah tertawa, dengan nilai sebelum tertawa sebesar 0,375 dB dan setelah tertawa 0,575 dB. Spektogram menunjukkan perubahan intensitas suara dominan pada rentang frekuensi hingga 1024 Hz selama 4 detik. Selain itu, analisis time domain dan power spectral density (PSD) menunjukkan peningkatan signifikan dalam frekuensi puncak dan intensitas suara setelah tertawa. Uji statistik menggunakan Shapiro-Wilk menunjukkan perbedaan signifikan secara statistik pada amplitudo suara setelah tertawa (nilai sig = 0,01), yang mengindikasikan pengaruh signifikan aktivitas tertawa terhadap karakteristik suara.

The paper is concerned with various algorithms used for the analysis based on the Discrete Wavelet Transform (DWT) of (non)stationary regimes involving almost sinusoidal waves and for data communication applications respectively. Different techniques are employed to perform analysis based on (de)compositions using two different trees: an unbalanced 10 level tree and a 6 level binary tree respectively. Our hybrid algorithms (for filters of length 4 and 6) are described and discussed. Their usability is demonstrated both for high and low power applications, a significant advantage being related to their fast “exact reconstruction” property. Considering the results and other important criteria (run time, memory consumptions), practical recommendations are made with respect to the selection of a proper reliable and fast DWT algorithm depending on the real applicability scenario. Keywords-discrete wavelet transform; hybrid algorithms; signals reconstruction; digital transmissions.

Agradezco y dedico esta tesis a: Mi esposa, por su amor y apoyo incondicional otorgado en todo momento y sobre todo para la culminación del trabajo. Mis padres, por ser en todo momento una fortaleza y ejemplo de vida y haber forjado en mí una persona con valores y metas. Mi hermano, por el cariño y la amistad que siempre me ha dado.

This project aims to do a study on the possibility of implementing an authentication application using biometrics in mobile devices. It has been carried out a brief review of existing methods of biometric recognition today and the possibilities offered by tablets and smartphones in this area. It was decided to use the built-in microphone all devices to collect and analyze the digital voice signal of the user and use this signal as a biometric identifier. It has developed a complete system of speaker verification with Matlab software tool. The system can be divided in two blocks. The first, processes audio signal in several stages: signal conditioning, windowing, segmenting voiced frames and formant extracting among other operations. Have been used two types of coefficients to characterize the speech. For side LPCC coefficients based on linear prediction filters, and secondly MFCC coefficients using the Fourier

In this paper we report the experiment carried out on recently collected speaker recognition database namely Arunachali Language Speech Database (ALS-DB)to make a comparative study on the performance of acoustic and prosodic features for speaker verification task.The speech database consists of speech data recorded from 200 speakers with Arunachali languages of NorthEast India as mother tongue. The collected database is evaluated using Gaussian mixture model-Universal Background Model (GMM-UBM) based speaker verification system. The acoustic feature considered in the present study is Mel-Frequency Cepstral Coefficients (MFCC) along with its derivatives.The performance of the system has been evaluated for both acoustic feature and prosodic feature individually as well as in combination.It has been observed that acoustic feature, when considered individually, provide better performance compared to prosodic features. However, if prosodic features are combined with acoustic feature, performance of the system outperforms both the systems where the features are considered individually. There is a nearly 5% improvement in recognition accuracy with respect to the system where acoustic features are considered individually and nearly 20% improvement with respect to the system where only prosodic features are considered.

In past few eras, emotion recognition from speech is one of the hottest research topic in the field of Human Computer Interaction. Many researches are going on various types of language, but for Bengali language, it is still very novice. In this work, 4 emotional state have been recognized i.e. happy, sad, angry and neutral from Bengali Speech Dataset. Proposed approach uses Pitch and Mel-frequency Cepstral Coefficient (MFCC) feature vectors to train k-Nearest Neighbor classifier for this work. A self-built Bengali emotional speech dataset has been used for both training and testing purpose. The dataset consists of consist of 50 people with 400 isolated emotional sentences. Using this dataset and above technique, we achieved 87.50% average accuracy rate, with detection accuracy each emotion (happy, sad, angry, neutral) respectively 80.00%, 75.00%, 85.00% and 75.00% in this work.

Resumen El guaraní es un idioma hablado por alrededor de ocho millones de personas. En Paraguay, cerca del 25 % de la población habla solamente guaraní. Con el incremento del número de dispositivos que incluyen asistentes personales controlados por voz, la necesidad de desarrollar aplicaciones con interfaces de voz está en crecimiento constante. A pesar de ser un idioma muy hablado, actualmente no existen investigaciones respecto al reconocimiento automático del habla en guaraní. En este contexto, proponemos un sistema de reconocimiento del habla en guaraní denominado Eñe'ẽ. Construimos el modelo utilizando las herramientas que ofrece el software de código abierto CMU Sphinx. Para ello, se recolectaron muestras de voz en guaraní, mediante el desarrollo de una aplicación web que permite a los interesados enviar lecturas de oraciones en guaraní que se muestran en pantalla. Este paper muestra los resultados de las pruebas experimentales que validan la eficiencia de la primera versión de Eñe'ẽ, y fomenta el desarrollo de sistemas utilizando este reconocedor de propósito general para futuras aplicaciones específicas.

Identifying characteristics of articulatory impairment in speech motor disorders is complicated due to the time-consuming nature of kinematic measures. The goal is to explore whether analysing the acoustic signal in terms of total squared changes of Mel-Frequency Cepstral Coefficients (TSC_MFCC) and its pattern over time provides sufficient spectral information to distinguish mild and moderate dysarthric French speakers with Amyotrophic Lateral Sclerosis (ALS) and Parkinson's Disease (PD) from each other and from healthy speakers. Participants produced the vowel-glide sequences /ajajaj/, /ujujuj/, and /wiwiwi/. From the time course of TSC_MFCCs, event-related and global measures were extracted to capture the degree of acoustic change and its variability. In addition, durational measures were obtained. For both mild and moderately impaired PD and ALS speakers, the degree of acoustic change and its variability, averaged over the complete contour, separated PD and ALS speakers from...

Category (2). Problems in voice production can appear due to functional disorders and laryngeal pathologies. The presence of laryngeal pathologies can causes significant changes in the vibrational patterns of the vocal folds and it is demonstrated that the impact of such pathologies can be reduced through continuous speech therapy. We propose a methodology based on non-parametric cepstral coefficients in Mel and Bark scales. The most relevant features are automatically selected using two algorithms, one is based on Principal Components Analysis (PCA) and other is based on Sequential Floating Features Selection (SFFS). In order to decide whether a voice recording is healthy or pathological, four different classifiers are implemented: linear and quadratic Bayesian, K nearest neighbors and Parzen. The best result was 89.18%, it was obtained from the union between MFCC and BFCC.

Online Social Network (OSN) adalah aplikasi social media yang memungkinkan komunikasi publik dan berbagi informasi. Namun, akun palsu di OSN dapat menyebarkan informasi palsu dengan sumber yang tidak diketahui. Ini adalah tugas yang menantang untuk mendeteksi akun berbahaya dalam sistem OSN yang besar. Keberadaan akun palsu atau akun yang tidak dikenal di OSN dapat menjadi masalah serius dalam menjaga privasi data. Berbagai komunitas telah mengusulkan banyak teknik untuk menangani akun palsu di OSN, termasuk teknik hitam-putih berbasis aturan hingga pendekatan pembelajaran. Oleh karena itu, dalam penelitian ini kami mengusulkan model klasifikasi menggunakan RNN untuk mendeteksi akun palsu secara akurat dan efektif. Kami melakukan penelitian ini dalam beberapa langkah, termasuk mengumpulkan dataset, pra-pemrosesan, ekstraksi, melatih model kami menggunakan RNN. Berdasarkan hasil eksperimen, model yang kami usulkan dapat menghasilkan akurasi yang lebih tinggi daripada model pembelajar...

Layanan hotline Universitas Pembangunan Nasional “Veteran” Yogyakarta merupakan layanan yang dapat digunakan oleh semua orang. Layanan tersebut digunakan dosen dan pegawai untuk berbagi informasi dengan bagian-bagian yang berlokasi di gedung rektorat. Penelepon dapat berkomunikasi dengan bagian yang dituju apabila telah teridentifikasi oleh petugas layanan hotline. Terminologi identitas yang terdiri dari nama, jabatan serta asal jurusan atau bagian ditanyakan saat proses identifikasi. Tidak terdapat catatan hasil identifikasi penelepon baik dalam bentuk fisik maupun basis data yang terekam pada komputer. Hal tersebut mengakibatkan tidak adanya dokumentasi yang dapat dijadikan barang bukti untuk menindak lanjuti kasus kesalahan identifikasi.  Penelitian ini fokus untuk mengurangi resiko kesalahan identifikasi penelepon menggunakan teknologi speaker recognition. Frekuensi suara diekstraksi menggunakan metode Mel-Frequency Cepstral Coefficient (MFCC) sehingga dihasilkan nilai Mel Frequ...

A Pattern can be characterized by more or less rich & varied pieces of information of different features. The fusion of these different sources of information can provide an opportunity to develop more efficient biometric system which is known as Multimodal biometric System. Multimodal biometrics is the combination of two or more modalities such as signature and speech modalities. In this work an offline signature verification system and speech verification system are combined as these modalities are widely accepted and natural to produce. This combination of multimodal enhances security and accuracy. In this work, database can be gathered from 14 users. Each user contributes 4 samples of signature & speech also. Forgeries are also added to test system. 14 forgeries are used for testing purpose. SIFT features are extracted for offline signature which results as a feature vector of 128 numbers & MFCC features are extracted for speech which results as a feature vector of 195 numbers. Fusion at feature extraction level is used in this work by using a new technique named msum which can be proposed by combining sum method & mean method. The experimental results demonstrated that the proposed multimodal biometric system achieves a recognition accuracy of 98.2% and with false rejection rate (FRR) of = 0.9% & false acceptance rate (FAR) of = 0.9%.

Speech is the most common form of communication, and the need of the hour is a robust speech recognition system. This paper aims to present an algorithm to design a continuous speech recognition system. The recognition of the speech utterances is done on a real-time basis using NI LabVIEW.

En este documento se desarrollan algunos métodos específicos para llevar a cabo un clasificador que sea capaz de identificar manatíes adultos y cría de distintos audios. La duración de las vocalizaciones se evaluó con un método de eliminación de ruido basado en eliminación de silencios, para luego aprovechar esta señal y evaluar uno de los principales parámetros, la frecuencia fundamental la cual se calculó con dos métodos (cepstrum y THD) y de esta manera se compararon los valores obtenidos y se evaluaron en un clasificador de mínima distancia.

En este artículo, investigamos el funcionamiento de distintos tipos de rasgos acústicos en un sistema de reconocimiento automático de habla (SRAH) en entorno telefónico. En concreto, exploramos dos alternativas distintas para el diseño del módulo parametrizador. En la primera de ellas, las características de dicho módulo son elegidas de forma empírica o basándose en conocimiento psicoacústico. En la segunda, dichas características son determinadas mediante la extracción discriminativa de rasgos que permiten una optimización conjunta del parametrizador y clasificador. Ambas estrategias han sido aplicadas a parametrizadores basados en la transformada ondicular dando lugar a mejoras significativas en la tasa de reconocimiento del sistema en comparación con las parametrizaciones convencionales basadas en la transformada de Fourier.

Este artículo presenta el desarrollo de un sistema de reconocimiento de palabras aisladas independiente del locutor, para comandar una silla de ruedas. Cada palabra se codifica mediante las técnicas de Predicción lineal y Cepstrum real, y la etapa de clasificación se realiza mediante Modelos Ocultos de Markov (HMM). A partir de los resultados se generan órdenes a un sistema móvil (silla de ruedas), el cual es comandado en un ambiente controlado.

In this study, speech to text system for homophone phrases in Indonesian was designed using an extraction method which featured Mel Frequency Cepstral Coefficient (MFCC). Feature extraction results were classified by comparing the two classifiers of Backpropagation Neural Network (BPNN) and K-Nearest Neigbour (KNN). The input data used were the recordings of each of 3 male and female respondents. The recording process was conducted for 5 seconds at a sampling frequency of 16 kHz and at channel mono. Classification results with test data to BPNN showed accuracy rates of 96.67% and 90% respectively for male and female respondents. Moreover, the level of accuracy obtained on KNN amounted to 83.33% for males and 73.33% for females.

The advent of modern communications and the low cost of some kinds of devices have resulted in a desire to equip elderly peoples' homes with sensors to monitor their activities and be forewarned of abnormal situations. In such an environment, sound may represent a rich source of information that can be exploited and this is considered as one of the most ergonomic and least intrusive solutions. However, this solution is often adversely affected by noise that is to say, mostly sounds of a type not taken into account in the creation of this system. Several methods were used to make it possible to classify sounds. In this work we tested Support Vector Machines to classify sounds in a domotic environment.

The use of face masks has increased dramatically since the COVID-19 pandemic startedin order to to curb the spread of the disease. Additionally, breakthrough infections caused by theDelta and Omicron variants have further increased the importance of wearing a face mask, even forvaccinated individuals. However, the use of face masks also induces attenuation in speech signals,and this change may impact speech processing technologies, e.g., automated speaker verification(ASV) and speech to text conversion. In this paper we examine Automatic Speaker Verification (ASV)systems against the speech samples in the presence of three different types of face mask: surgical,cloth, and filtered N95, and analyze the impact on acoustics and other factors. In addition, weexplore the effect of different microphones, and distance from the microphone, and the impact offace masks when speakers use ASV systems in real-world scenarios. Our analysis shows a significant deterioration in performance when an ASV system encounters different face masks, microphones, and variable distance between the subject and microphone. To address this problem, this paper proposes a novel framework to overcome performance degradation in these scenarios by realigning the ASV system. The novelty of the proposed ASV framework is as follows: first, we propose a fused feature descriptor by concatenating the novel Ternary Deviated overlapping Patterns (TDoP), Mel Frequency Cepstral Coefficients (MFCC), and Gammatone Cepstral Coefficients (GTCC), which are used by both the ensemble learning-based ASV and anomaly detection system in the proposed ASV  architecture. Second, this paper proposes an anomaly detection model for identifying vocal samples produced in the presence of face masks. Next, it presents a Peak Norm (PN) filter to approximate the signal of the speaker without a face mask in order to boost the accuracy of ASV systems. Finally, the features of filtered samples utilizing the PN filter and samples without face masks are passed to the proposed ASV to test for improved accuracy. The proposed ASV system achieved an accuracy of 0.99 and 0.92, respectively, on samples recorded without a face mask and with different face masks. Although the use of face masks affects the ASV system, the PN filtering solution overcomes this deficiency up to 4%. Similarly, when exposed to different microphones and distances, the PN approach enhanced system accuracy by up to 7% and 9%, respectively. The results demonstrate the effectiveness of the presented framework against an in-house prepared, diverse Multi Speaker Face Masks (MSFM) dataset, (IRB No. FY2021-83), consisting of samples of subjects taken with a variety of face masks and microphones, and from different distances.

In this paper, we focused on automation of Dunstan Baby Language. This system uses MFCC as feature extraction and codebook as feature matching. The codebook of clusters is made from the proceeds of all the baby's cries data, by using the k-means clustering. The data is taken from Dunstan Baby Language videos that has been processed. The data is divided into two, training data and testing data. There are 140 training data, each of which represents the 28 hungry infant cries, 28 sleepy infant cries, 28 wanted to burp infant cries, 28 in pain infant cries, and 28 uncomfortable infant cries (could be because his diaper is wet/too hot/cold air or anything else). The testing data is 35, respectively 7 infant cries for each type of infant cry. The research varying frame length: 25 ms/frame length = 275, 40 ms/frame length = 440, 60 ms/ frame length = 660, overlap frame: 0%, 25%, 40%, the number of codewords: 1 to 18, except for frame length 275 and overlap frame = 0 using 1 to 29 clusters. The identification of this type of infant cries uses the minimum distance of euclidean distance. Accuracy value is between 37% and 94%. Sound 'eh' is the most familiar, whereas sound 'owh' is always missunderstood and generally it is known as 'neh' and 'eairh'. The weakness point of this research is the silent is only be cut at the beginning and at the end of speech signal. Hopefully, in the next research, the silent can be cut in the middle of sound so that it can produce more specific sound. It has impact on the bigger accuracy as well.

English fricatives’ pronunciation is problematic to common people especially Indo-European speakers such as Norwegian, French, Slovakian, Slovenian, Bulgarian, Portuguese, Czech, Spanish, Dutch, and Greek in the research. This research is about a problem of pronouncing English fricatives, which is experienced by speakers of Indo-European language. The theory used for this research is the availability of 9 English fricatives, mentioned by Peter Roach, in Indo-European language and how they affect the language. This research has 2 methods, which are data collection by recording the Indo-European speakers, then giving a questionnaire afterward and analyzing the data collection’s result. The results show 94% Dutch speakers achieved the most correct pronunciation of fricatives, the most fricatives, which were always replaced, are dental and palato-alveolar fricatives, and how English fricatives did not surround their daily activities back in their countries.

En este trabajo se presenta un sistema de reconocimiento de palabras aisladas dependiente del locutor. Cada palabra se codifica mediante las técnicas de Predicción Lineal y Cepstrum real, mientras que la etapa de clasificación se realiza mediante el alineamiento temporal dinámico, que permite independencia del intervalo de tiempo de cada muestra de voz. Los resultados obtenidos demuestran que el uso de estas técnicas permiten obtener un 85% de clasificación correcta. Palabras claves: Reconocimiento de voz, análisis Cepstral, extracción de parámetros, patrones de voz.

Speaker recognition is the identification of the person who is speaking by characteristics of their voices, also called "voice recognition". The components of Speaker Recognition includes Speaker Identification(SI) and Speaker Verification(SV). Speaker identification is the task of determining an unknown speakers identity. If the speaker claims to be of a certain identity and the voice is to verify this claim, this is called Speaker Verification. It determines whether an unknown voice matches the known voice of a speaker whose identity is being claimed. This paper proposes Speaker Verification task. There are two phases in the Speaker Verification task namely, training and testing. In the training phase, different features such as Mel Frequency Cepstral Coefficient(MFCC), Linear Predictive Cepstral Coefficient(LPCC), Perceptual Linear Predictive(PLP) are extracted from the speech signal and is trained by Support Vector Machine to get the target speaker model. It is trained with both actual speaker and impostor utterances. In the testing phase, features are extracted from the test speech signal. The test features are extracted for different duration of time. The extracted feature vectors are given to the claimed speaker model and the decision is taken as authorised speaker or an impostor. The performance of a speaker verification task is analysed using different features with different utterance sizes. The result shows that the performance of a speaker verification task decreases when the duration of the speech utterances decreased.

Tindak pidana yang dilakukan pelaku kejahatan sedikit tidaknya terdapat barang bukti digital yang ditinggalkan berupa rekaman suara yang dihasilkan dari percakapan menggunakan telepon, algoritma dalam menganalisis suara rekaman banyak beredar di internet salah satunya algoritma KNN yang biasanya digunakan untuk klasifikasi, identifikasi dan prediksi. Penelitian ini menggunakan algoritma tersebut dengan metode Minkowski untuk melakukan pengenalan sampel suara rekaman percakapan pelaku kejahatan dengan sampel suara tersangka. Melibatkan dua responden yang berperan sebagai tersangka dan tiga responden berperan sebagai pelaku. Masing-masing responden akan melakukan dubbing (perekaman suara) dan dipotong menjadi sebelas bagian sebagai data latih dan data uji. sampel suara percakapan akan di extract menggunakan teknik MFCC untuk mendapatkan nilai spectrum yang selanjutnya akan diproses menggunakan algoritma KNN dan metode minkowski menggunakan Python. Hasil dari penelitian ini adalah dari...

The ECG Signal is one of the most important biomedical signals for diagnostic of different pathologies, and for that reason it is necessary to look for new methods of compression of the signal in search of a fast transmission and best storage without losing his quality. In the ...

Emotion recognition (ER) from speech signals is a robust approach since it cannot be imitated like facial expression or text based sentiment analysis. Valuable information underlying the emotions are significant for human-computer interactions enabling intelligent machines to interact with sensitivity in the real world. Previous ER studies through speech signal processing have focused exclusively on associations between different signal mode decomposition methods and hidden informative features. However, improper decomposition parameter selections lead to informative signal component losses due to mode duplicating and mixing. In contrast, the current study proposes VGG-optiVMD, an empowered variational mode decomposition algorithm, to distinguish meaningful speech features and automatically select the number of decomposed modes and optimum balancing parameter for the data fidelity constraint by assessing their effects on the VGG16 flattening output layer. Various feature vectors were employed to train the VGG16 network on different databases and assess VGG-optiVMD reproducibility and reliability. One, two, and three-dimensional feature vectors were constructed by concatenating Mel-frequency cepstral coefficients, Chromagram, Mel spectrograms, Tonnetz diagrams, and spectral centroids. Results confirmed a synergistic relationship between the fine-tuning of the signal sample rate and decomposition parameters with classification accuracy, achieving state-of-the-art 96.09% accuracy in predicting seven emotions on the Berlin EMO-DB database.

Приведены результаты оценки компетенций персонала, полученные на основе нейросетевого подхода. Термин «компетенции» используется во многих научных дисциплинах и практических приложениях. В управлении персоналом компетенция представляет собой формально описанные требования к профессиональным качествам работника. Важность компетенций подчеркивается в международных стандартах ИСО 9001 и моделях национальных и транснациональных премий в области качества продукции. Однако при оценивании компетенций окончательной точки зрения еще не сформировано. Существуют различные методы и приемы, позволяющие с тех или иных позиций получить оценку компетенций сотрудников, которая сводится, чаще всего, к субъективному подходу. В работе для оценки компетенций предложено использовать искусственные нейронные сети, с их помощью показана возможность классификации работников по уровню компетенций, приведены результаты моделирования работы нейронной сети с помощью инструмента Simulink. ФИНАНСИрОВАНИЕ. Исследование выполнено при финансовой поддержке РФФИ в рамках научного проекта № 18-010-00338А. КЛЮЧЕВЫЕ СЛОВА: компетенция работника, оценка компетенций, нейронная сеть, имитация работы системы оценки.

A Pattern can be characterized by more or less rich & varied pieces of information of different features. The fusion of these different sources of information can provide an opportunity to develop more efficient biometric system which is known as Multimodal biometric System. Multimodal biometrics is the combination of two or more modalities such as signature and speech modalities. In this work an offline signature verification system and speech verification system are combined as these modalities are widely accepted and natural to produce. This combination of multimodal enhances security and accuracy. In this work, database can be gathered from 14 users. Each user contributes 4 samples of signature & speech also. Forgeries are also added to test system. 14 forgeries are used for testing purpose. SIFT features are extracted for offline signature which results as a feature vector of 128 numbers & MFCC features are extracted for speech which results as a feature vector of 195 numbers. Fusion at feature extraction level is used in this work by using a new technique named msum which can be proposed by combining sum method & mean method. The experimental results demonstrated that the proposed multimodal biometric system achieves a recognition accuracy of 98.2% and with false rejection rate (FRR) of = 0.9% & false acceptance rate (FAR) of = 0.9%.

En este artículo se describe el sistema de reconocimiento de locutor implementado por el I3A para la evaluación del NIST 2008. Se dispone de dos sistemas básicos: GMM-UBM likelihood ratio y GMM-SVM. Las señales proporcionadas por el NIST para la evaluación han sido adquiridas a través de diferentes micrófonos y canales de comunicación, se discutirá como afectan las diferentes técnicas de compensación de canal al funcionamiento del sistema. Se presentan los resultados obtenidos durante el desarrollo del sistema sobre la base de datos de 2006 y los obtenidos en 2008.

In this paper, we present an approach to develop an automatic speech recognition (ASR) system of Urdu isolated words. Our experimentation is based on a medium vocabulary speech corpus of Urdu, consisting of 250 words. We develop our approach using the open source Sphinx toolkit. Using this platform, we extract the Mel Frequency Cepstral Coefficients (MFCC) features and build a Hidden Markov Model to perform recognition task. We report percentage accuracy for two different experiments based on 100 and 250 words respectively. Experimental results suggest that better recognition accuracy has been achieved with this approach, as compared to the previous results reported on this corpus.

Indonesia merupakan negara besar yang memiliki banyak keberagaman budaya dan suku sehingga memiliki banyak bahasa atau dialek yang berbeda-beda di setiap daerah. Berbagai penelitian dalam pengolahan sinyal suara telah banyak dikembangkan. Salah satu penelitian yang menarik untuk dikembangkan adalah Identifikasi Dialek. Dalam penelitian ini, dibuat suatu program atau aplikasi Speech Recognition dengan metode Mel-Frequency Ceptrums Coefficients (MFCC) dan Adaptive Neuro Fuzzy Inference System (ANFIS) untuk Bahasa Jawa dialek Betawi, Sunda, Banyumasan, dan Suroboyoan. MFCC untuk proses ekstraksi ciri dari sinyal ucapan pembicara dimana prosesnya akan mengubah sinyal suara menjadi beberapa vektor ciri yang kemudian akan ditampilkan dalam bentuk grafik. Analisis dan perancangan bentuk pola suara menggunakan software Phyton. Pengujian dilakukan dengan cara merancang perangkat lunak, melakukan pengambilan data, training sistem ANFIS, pembuatan GUI, dan pengujian dalam pengolahan data pengu...

In this work, a 5 state left to right HMM-based Bangla Isolated word speech recognizer has been developed. To train and test the recognizer, a small corpus of various sampling frequencies have been developed in noisy as well as the noiseless environment. The number of filter banks is varied during the feature extraction phase for both MFCC and PLP. The effects of 2nd and 3rd differential coefficients have also been observed. Experimental results exhibit that MFCC based feature extraction technique is better in CLASSROOM environment on the contrary PLP based technique performs better not only in a noiseless environment but also in when AC or FAN noise is present. We have also noticed that higher sampling frequency and higher filter order don't always help to improve the performance.

Layanan hotline Universitas Pembangunan Nasional “Veteran” Yogyakarta merupakan layanan yang dapat digunakan oleh semua orang. Layanan tersebut digunakan dosen dan pegawai untuk berbagi informasi dengan bagian-bagian yang berlokasi di gedung rektorat. Penelepon dapat berkomunikasi dengan bagian yang dituju apabila telah teridentifikasi oleh petugas layanan hotline. Terminologi identitas yang terdiri dari nama, jabatan serta asal jurusan atau bagian ditanyakan saat proses identifikasi. Tidak terdapat catatan hasil identifikasi penelepon baik dalam bentuk fisik maupun basis data yang terekam pada komputer. Hal tersebut mengakibatkan tidak adanya dokumentasi yang dapat dijadikan barang bukti untuk menindak lanjuti kasus kesalahan identifikasi.  Penelitian ini fokus untuk mengurangi resiko kesalahan identifikasi penelepon menggunakan teknologi speaker recognition. Frekuensi suara diekstraksi menggunakan metode Mel-Frequency Cepstral Coefficient (MFCC) sehingga dihasilkan nilai Mel Frequ...

A phoneme is basically a very small number of basic sounds from any human spoken language. Many researches have been conducted in ASR from last few centuries. Phoneme recognition with 40 native speakers is taken into consideration with MFCC, LFs and HMM where MFCC provides more accuracy than LF methods [1]. READ is a phoneme recognition system proposed in West Bengal (India) for the development of a Bangla ASR claimed to provide 98.35% accuracy of Bangla vowel
phonemes with the their used speech samples. But Indian (West Bengal) Bangla accent differs little from Bangladeshi (Bangladesh) Bangla accent [2]. This could be an issue for this research. In another research, HMM with single and multilayer neural network is taken into consideration for Bangla phoneme recognition [3]. Local feature (LF)-based automatic speech recognition (ASR) system provides higher phoneme accuracy rate than the MFCC-based system [4]. An experiment for Bangla phoneme recognition noticed that changing of window length (Hamming window frame from 20 to 64 ms duration) during framing slightly influences the
recognition rate [5]. Very few researches are conducted with in this area especially in Bangla phoneme recognition. So, more research needs to be conducted for Bangla phoneme recognition.

Online Social Network (OSN) adalah aplikasi social media yang memungkinkan komunikasi publik dan berbagi informasi. Namun, akun palsu di OSN dapat menyebarkan informasi palsu dengan sumber yang tidak diketahui. Ini adalah tugas yang menantang untuk mendeteksi akun berbahaya dalam sistem OSN yang besar. Keberadaan akun palsu atau akun yang tidak dikenal di OSN dapat menjadi masalah serius dalam menjaga privasi data. Berbagai komunitas telah mengusulkan banyak teknik untuk menangani akun palsu di OSN, termasuk teknik hitam-putih berbasis aturan hingga pendekatan pembelajaran. Oleh karena itu, dalam penelitian ini kami mengusulkan model klasifikasi menggunakan RNN untuk mendeteksi akun palsu secara akurat dan efektif. Kami melakukan penelitian ini dalam beberapa langkah, termasuk mengumpulkan dataset, pra-pemrosesan, ekstraksi, melatih model kami menggunakan RNN. Berdasarkan hasil eksperimen, model yang kami usulkan dapat menghasilkan akurasi yang lebih tinggi daripada model pembelajar...

In past few eras, emotion recognition from speech is one of the hottest research topic in the field of Human Computer Interaction. Many researches are going on various types of language, but for Bengali language, it is still very novice. In this work, 4 emotional state have been recognized i.e. happy, sad, angry and neutral from Bengali Speech Dataset. Proposed approach uses Pitch and Mel-frequency Cepstral Coefficient (MFCC) feature vectors to train k-Nearest Neighbor classifier for this work. A self-built Bengali emotional speech dataset has been used for both training and testing purpose. The dataset consists of consist of 50 people with 400 isolated emotional sentences. Using this dataset and above technique, we achieved 87.50% average accuracy rate, with detection accuracy each emotion (happy, sad, angry, neutral) respectively 80.00%, 75.00%, 85.00% and 75.00% in this work.

In past few eras, emotion recognition from speech is one of the hottest research topic in the field of Human Computer Interaction. Many researches are going on various types of language, but for Bengali language, it is still very novice. In this work, 4 emotional state have been recognized i.e. happy, sad, angry and neutral from Bengali Speech Dataset. Proposed approach uses Pitch and Mel-frequency Cepstral Coefficient (MFCC) feature vectors to train k-Nearest Neighbor classifier for this work. A self-built Bengali emotional speech dataset has been used for both training and testing purpose. The dataset consists of consist of 50 people with 400 isolated emotional sentences. Using this dataset and above technique, we achieved 87.50% average accuracy rate, with detection accuracy each emotion (happy, sad, angry, neutral) respectively 80.00%, 75.00%, 85.00% and 75.00% in this work.

This study aims to analyze the characteristics of Japanese sibilant sound observed from the acoustic
analysis. This study's data were in the form of sound samples from 16 respondents who spoke
Javanese in Japanese language learners at the university and 8 Japanese native speakers. The
respondents were asked to pronounce vocabulary containing Japanese sibilant sounds at the
beginning and middle of 72 words that have meaning. The sample of Indonesian and Japanese
respondent's sound was measured using Praat software to find their duration, intensity, and Center
of Gravity (CoG) of the sibilant sound. The measurement results are carried out an analysis of 2
factors, namely consonant type, and sound position, using ANOVA statistics. The duration,
intensity, and Center of Gravity measurement results show that the two factors (consonant type and
sound position) are statistically significant. The distribution of sibilant sound duration is from 68
ms to 144 ms, and the distribution of sibilant sound intensity is from 50 dB to 60 dB

In these modern times there are several methods of communication between ourselves and even our machines, these allow us to create ties between people contact or via an interface for human-machine interaction. These techniques have been developed over time by the constant evolution of technology, triggered by a futuristic world and wants to be globalized. One of these techniques has been developed, is the recognition of Subvocal also known speech or silent speech, which allows us to analyze the signals from the vocal apparatus without the need for sound waves or pronounce words.

Agradezco y dedico esta tesis a: Mi esposa, por su amor y apoyo incondicional otorgado en todo momento y sobre todo para la culminación del trabajo. Mis padres, por ser en todo momento una fortaleza y ejemplo de vida y haber forjado en mí una persona con valores y metas. Mi hermano, por el cariño y la amistad que siempre me ha dado.

There are a number of languages accent differential applications that detect the different accents in assorted languages. The studies which have done before most of them are based on the English language and different languages throughout the world. A few researches have been performed in Bangla regional language accent differential applications, which is not conclusive for the system to be able to manage Bangla accented speakers. In this paper, we report regional language accent detection experiments of different types of Bangladesh. We demonstrate a strategy to observe Bangladeshi different accents which exploit Mel frequency cepstral coefficient (MFCC) and recurrent neural network (RNN). Listening from the people of different places in Bangladesh creates an accent differentiation results performed by the speakers. This experimental result shows the adaptation of the people to adapt of the regional languages. Keywords Accent differential application • MFCC • Recurrent neural network •

El objetivo de esta comunicacion es presentar las actividades realizadas desde noviembre de 1994 dentro del proyecto “Speaker Recognition in Telephony”, financiado por la Comunidad Europea en el marco del programa “European Cooperation in the Field of Scientific and Technical Research” (COST), con referencia COST250. Los paises participantes en el proyecto COST250 son Belgica, Dinamarca, Eslovenia, Espana, Francia, Grecia, Inglaterra, Irlanda, Italia, Holanda, Portugal, Suecia, Suiza y Turquia.