Santanu Phadikar

Line spectral frequency-based features and extreme learning machine for voice activity detection from audio signal

Voice activity detection (VAD) refers to the task of identifying vocal segments from an audio clip. It helps in reducing the computational overhead as well elevate the recognition performance of speech-based systems by helping to discard the non vocal portions from an input signal. In this paper, a VAD technique is presented that uses line spectral frequency-based statistical features namely LSF-S coupled with extreme learning-based classification. The experiments were performed on a database of more than 350 h consisting of data from multifarious sources. We have obtained an encouraging overall accuracy of 99.43%.

REARC-a Bangla Phoneme recognizer

The field of Information Technology has undergone an evolution in the past couple of decades and so has risen the need for development of better modes of interaction between the users and the electronic devices for exhaustive exploration of the world of IT with minimum hindrance. Speech can be one such mode of interaction as this has been the primary mode of communication since the advent of human civilization. Speech Recognition is the technique of identification of uttered words from voice signals. Every language is composed of an inventory of sounds called Phoneme set which makes up its entire vocabulary. Speech Recognition in Bangla is rather a complicated task due to the presence of compound characters and complex nature of the language itself. REARC (Record Extract Approximate Reduce Classify) is a Bangla Phoneme Recognition system which is designed to pave the way for a Bangla Speech Recognizer. At the outset, Mel Scale Cepstral Coefficient (MFCC) features have been used for the system on a database of 3150 Bangla Vowel Phonemes and an accuracy of 98.22% has been obtained.

MISNA - A musical instrument segregation system from noisy audio with LPCC-S features and extreme learning

Technology has developed a lot over the last decades and has made a profound impact in almost every field. The field of Music Information Retrieval (MIR) has not been an exception to this as well, one of its most promising applications being Automatic Music Transcription (AMT). It is important to identify the active regions of various Instruments in a piece before transcription and the challenge elevates even more when the audio clips are contaminated with noise. MISNA (Musical Instrument Segregation from Noisy Clips) is a system proposed towards the identification of isolated Instruments from noisy clips which can aid towards AMT in noisy environments. The system works using statistical features (LPCC-S) derived from raw Linear Predictive Cepstral Coefficient values on very short clips of lengths 1 and 2 seconds. The system has been tested for various SNR scenarios and highest accuracies of 98.63% and 97.42% for Individual Instruments and Instrument Family identification has been obtained with the aid of Extreme Learning based classifier for a highest of 2626 clips.

Log-Based Cloud Forensic Techniques: A Comparative Study

Cloud computing is one of the most recent advancements in the field of distributed computing. It has gained a lot of attention due to its on demand, pay-per-use service, and all time availability, reliability, and scalability. Although it offers numerous advantages, but due to its multi-tenant architecture, it is prone to various malicious attacks and illegal activities. Cloud service provider (CSP) takes the responsibility to secure customers’ data against such attacks. In the event of such malicious activities, CSP aims to trace the intruder. Cloud forensic techniques help in identifying the attacker along with proper evidence in cloud platform. Components of clouds such as log records are then analyzed to track for such detrimental activities. In this paper, some existing log-based cloud forensic techniques have been widely studied. The detailed comparative analysis has been done for the various techniques based on their advantages and limitations. By exploring the limitations and advantages of the existing approaches, future research areas have been identified.

READ—A Bangla Phoneme Recognition System

Speech Recognition is a challenging task especially for a multilingual country like India as the speakers are habituated in using mixed language and accent. Bangla is a very popular language in East Asia and a fully functional Automated Speech Recognition System (ASR) for it is yet to be developed. Every language embodies a set of sounds called phoneme set, which is the building block for the words of that language. READ (Record Extract Approximate Distinguish) is a Bangla phoneme recognition system, proposed toward the development of a Bangla ASR. To start with, Mel Scale Cepstral Coefficient (MFCC) features have been used for testing on a database of 1400 Bangla vowel phonemes and an accuracy of 98.35% has been obtained.

A Cloud Intrusion Detection System Using Novel PRFCM Clustering and KNN Based Dempster-Shafer Rule

Cloud computing has established a new horizon in the field of Information Technology. Due to the large number of users and extensive utilization, the Cloud computing paradigm attracts intruders who exploit its vulnerabilities. To secure the Cloud environment from such intruders an Intrusion Detection System IDS is required. In this paper the authors have proposed an anomaly based IDS which classifies an incoming connection by taking the deviation of it from the normal behaviors. The proposed method uses a novel Penalty Reward based Fuzzy C-Means PRFCM clustering algorithm to generate a rule set and the best rule set is extracted from it using a modified approach for KNN algorithm. This best rule set is used in evidential reasoning of Dempster Shafer Theory for classification. The IDS has been trained and tested with NSL-KDD dataset for performance evaluation. The results prove the proposed IDS to be highly efficient and reliable.

Importance of Thermal Features in the Evaluation of Bacterial Blight in Rice Plant

The aim of the study is to investigate the potential of texture and thermal features extracted from the thermograph of the rice leaves in bacterial leaf blight forecasting. Thermal images have some advantages over visual images. Visual images are capable of capturing only symptoms visible in bare eyes whereas thermal images can capture invasive temperature changes of an object when any chemical changes occur within it, which may not create any visual changes. In this paper, thermal images are used to identify the internal changes of the rice leaves before any visual changes occur due to the bacterial leaf blight disease. Thermal images of the leaves at normal, primary stage of infection and highly infected stage are collected from field. For this experiment, 158 samples of each stage (normal leaves, leaves at primary stage of infection and highly infected leaves) are considered. Images are preprocessed to standardize the environment of image acquisition. Then images are segmented to extract the region of interest using Otsu’s algorithm. Temperature variation and texture features are extracted from the segmented images using the Flir Tools and Gray-level co-occurrence matrix method respectively. The temperature differences of normal leaves, leaves at primary stage of infection and highly infected leaves are evaluated using summary statistics. Paired t-test values are computed to find the significance of the result. The result shows that there is significant difference among these three stages of leaves with respect to thermal feature. But, with respect to the texture features there is no significant difference. Hence, the result verifies the importance of thermal features in bacterial leaf blight forecasting.

An Ensemble Learning Based Bangla Phoneme Identification System Using LSF-G Features

Technology has evolved a lot in the last decade, and various devices have come up for assisting us in our day-to-day life. There has always been a need for simplifying the User Interfaces (UI) of such devices so that they can be easily interacted with, and a speech based UI can be a potential solution. Speech recognition is the task of identification of words from voice signals. Every language consists of a set of atomic sounds called Phonemes which builds up the entire vocabulary of that language. Speech recognition in Bangla is rather a complicated task due to the complex nature of the language like the presence of compound characters. In this paper, a Bangla Phoneme recognition system is proposed towards the development of a Bangla Speech recognition system based on Line Spectral Frequency-Grade (LSF-G) features derived from standard line spectral frequency values. The system has been tested on a Bangla Swarabarna Phoneme dataset of 3290 clips and an accuracy of 94.01% has been obtained with an Ensemble learning based approach.

An Ensemble Learning-Based Bangla Phoneme Recognition System Using LPCC-2 Features

An array of devices have emerged lately for easing our daily life but one concern has always been towards designing simple user interface (UI) for such devices. A speech-based UI can be a solution to this, considering the fact that it is one of the most spontaneous and natural modes of interaction for most people. The process of identification of words and phrases from voice signals is known as Speech Recognition. Every language encompasses a unique set of atomic sounds termed as Phonemes. It is these sounds which constitute the vocabulary of that language. Speech Recognition in Bangla is a bit complicated task mostly due to the presence of compound characters. In this paper, a Bangla Phoneme Recognition system is proposed to help in the development of a Bangla Speech Recognizer using a new Linear Predictive Cepstral Coefficient-based feature, namely LPCC-2. The system has been tested on a data set of 3710 Bangla Swarabarna (Vowel) Phonemes, and an accuracy of 99.06% has been obtained using Ensemble Learning.

Segregation of Speech and Songs - A Precursor to Audio Interactive Applications

Audio interactive applications have eased our lives in numerous ways encompassing speech recognition to song identification. Such applications have helped the common people in using Information Technology by providing them a passage for skipping the complicated user interactivity procedures. Audio based search applications have become very popular nowadays especially for searching songs. A system which can distinguish between speech and songs can help to boost the performance of such applications by minimizing the search space and at the same time decide the method of recognition based on the type of audio. It can also help in music-speech separation from audio for karaoke development. In this paper, a system to segregate songs and speech has been proposed using Line Spectral Pair based features. The system has been tested on a database of 19374 clips and a highest accuracy of 99.88% has been obtained with Ensemble Learning based classification.