Vilas M. Thakare

Noise Estimation and Noise Removal Techniques for Speech Recognition in Adverse Environment

Noise is ubiquitous in almost all acoustic environments. The speech signal, that is recorded by a microphone is generally infected by noise originating from various sources. Such contamination can change the characteristics of the speech signals and degrade the speech quality and intelligibility, thereby causing significant harm to human-to-machine communication systems.

Domain Independent Approach for Aspect Oriented Sentiment Analysis for Product Reviews

The Sentiment analysis from text documents is emerging field for the research in Natural Language Processing (NLP) and text mining. Feature specific opinion matters more than the overall opinion. Given a collection of review texts, the goal is to detect the individual product aspects comments by reviewers and to decide whether the comments are rather positive or negative. In this research paper unsupervised approach for domain independent feature specific sentiment analysis has been proposed. SentiWordNet lexical resource has been used to determine the polarity of identified features. Research work has shown the promising results over the previously used approaches using SentiWordNet. Newly introduced SentiWordNet 3.0 has been proved to be an important lexical resource.

Feature Extraction for a Speech Recognition System in Noisy Environment: A Study

Speech feature extraction has been a key focus in robust speech recognition research. Selecting proper features is the key of effective system performance. Robustness to additive noise remains a large unsolved problem in automatic speech recognition research today. One of the environmental changes that have a large impact on the performance of current ASR systems is background noise. There are several approaches that one can take to improve ASR systems robustness [7, 10] to changes in background noise. One of these approaches is to address the problem at the feature extraction stage of the system. That is, to use a speech feature extraction algorithm that produces features that are as invariant as possible to background noise changes, while simultaneously capturing the salient speech information. Many feature extraction algorithms have been proposed that are designed specifically to have a low sensitivity to background noise. In this paper we are presenting some feature extraction algorithm developed for noisy environment.

Speech user interface for computer based education system

Speech is a very natural and basic way in human-to-human communication.

Performance Analysis of Various Feature Detector and Descriptor for Real-Time Video based Face Tracking

This paper presents the performance analysis of various contemporary feature detector and descriptor pair for real time face tracking. These feature detectors/descriptors are mostly used in image matching applications. Some feature detectors/descriptors like STAR, FAST, BRIEF, FREAK, and ORB can also be used for SLAM applications due to their high performance. However using only one of these feature detectors for object tracking may not provide good accuracy due to various challenges in tracking like abrupt change in object motion, non-rigid object structure, change in appearance of object, occlusions in the scene and camera motion. But it can be combined other object tracking algorithm to improve the overall tracking accuracy. In this paper we have measured the tracking speed and accuracy of these feature detectors in real time video for face tracking using parameters like average number of detected key points, average detection time of key-point, frame per second and number of matches using OpenCV. General Terms Object tracking, Image matching.

Digital image forensics: An affine transform robust copy-paste tampering detection

It became very easy today to capture and create digital photographs. Its no more a costlier affair, as most of the handheld electronic gadgets such as mobile phones are equipped with digital cameras. Today, there are ample PC and mobile apps available which are developed to manipulate captured photographs. One can easily take a picture, manipulate it with the installed app and make it viral through the internet. Hence, these digital photographs should not be interpreted as they speak. Digital images are the good proof of events and places. Hence, these digital photographs can be presented as evidence before a court of law. It becomes very important in such cases then, to prove the digital photographs in question to be original. Digital image forensics plays a vital role in such circumstances. Digital image forensics is a branch of digital forensics which deals with examining the digital photographs for their integrity and authenticity. In this paper, we present a digital image forensic method which can detect one of such image tampering. As images can be tampered in a number of ways, in this paper, we address a common case called as copy-paste tampering. Our proposed method is robust to affine transform; especially to rotation and scaling.

Dynamic Route and Threshold Level Selection in Energy Efficient WSN

Sensor nodes are hardware devices and their source of energy is battery power . Nodes store, forward, report various environment related parameters to the sink, which is normally a base station, thus monitor the environment using co-operative information. A dynamic approach is suggested to regulate and decide broadcast radius . It can reduce collision . It can improve life span of network path and nodes on the path . even if the tasks are scheduled at the speed, still there will be the idle intervals. These idle intervals arise due to two reasons: 1) the idle intervals that are inherent to fixed priority schedules and 2) the idle intervals that arise due to run-time variations in execution time of tasks. All the tasks that are ready are kept in the “ready queue” in the order of their priority. The task that is currently being executed is called “active-task”. At layer 1, consumption of energy is more . At layer 2, consumption of energy is slightly less . At layer 3, consumption of energy is low . Thus, Adjusting the operation mode of the nodes in wireless sensor networks can extend the networks lifetime effectively.

A Hybrid Method for Automatic Speech Recognition Performance Improvement in Real World Noisy Environment

It is a well known fact that, speech recognition sy stems perform well when the system is used in condi tions similar to the one used to train the acoustic model s. However, mismatches degrade the performance. In adverse environment, it is very difficult to predic t the category of noise in advance in case of real world environmental noise and difficult to achieve enviro nmental robustness. After doing rigorous experiment al study it is observed that, a unique method is not a vailable that will clean the noisy speech as well a s preserve the quality which have been corrupted by r eal natural environmental (mixed) noise. It is also observed that only back-end techniques are not suff icient to improve the performance of a speech recognition system. It is necessary to implement pe rformance improvement techniques at every step of back-end as well as front-end of the Automatic Spee ch Recognition (ASR) model. Current recognition systems solve this problem using a technique called adaptation. This study presents an experimental st udy that aims two points, first is to implement the hyb rid method that will take care of clarifying the sp eech signal as much as possible with all combinations of filters and enhancement techniques. The second poi nt is to develop a method for training all categories of noise that can adapt the acoustic models for a new environment that will help to improve the performan ce of the speech recognizer under real world environmental mismatched conditions. This experiment confirms that hybrid adaptation methods improve the ASR performance on both levels, (Signal-to-Noise Ratio) SNR improvement as well as word recognition accuracy in real world noisy environment.

Normalization techniques for hiding speakers identity

In most security domain transmission of secret and confidential voice data requires that the identity of the speaker should not get revealed but at the same time data contents in speech signal should be interpreted. Speech and speaker normalization techniques help to solve this issue. This paper proposes speaker normalization technique that attempts to alter the characteristics of speakers voice so that it would be difficult for an eavesdropper to identify the speaker. Intended listener can retrieve the particular information. Evaluations of subjective and objective tests confirm the effectiveness of the proposed method and because of this method security enhances.

Improving speech intelligibility in an adverse condition using subband spectral subtraction method

Many people have great difficulty in Understanding speech with background noise. Speech Enhancement plays a vital role in such situations. The background noise has to be removed from the noisy speech signal to increase the signal intelligibility and to reduce the listener fatigue. In this paper, a novel approach is used to enhance the perceived quality of the speech signal when the additive noise cannot be directly controlled. The proposed approach is a speech enhancement method based on the preprocessed sub band spectral subtraction method, and the preprocessing is done by using partial differential equation.