Prof. K. T. Talele

Speech recognition has found its application on various aspects of our daily lives from automatic phone answering service to dictating text and issuing voice commands to computers. In this paper, we present the steps involved in the design of a speaker-independent speech translation system. We focus mainly on the pre-processing stage that extracts salient features of a speech signal and a technique called Dynamic Time Warping commonly used to compare the feature vectors of speech signals. These techniques are applied for recognition of isolated words spoken. Finally, we design a simple Voice-to-Voice converter application using independent hardware system.

Spherical robots utilize an original locomotion system by displacing its center of gravity in order to generate torque and rotate itself. Spherical robots, however, can generate only small amount of torque for rotation. This makes its locomotion ability for uphill climbing and obstacle over running greatly limited. Therefore, this study focuses on spherical robots operating on rough and uneven floors. It describes the basic dynamics and control methods, which can applied for enhancing the features and overcoming the disadvantages.

The paper describes the project that is intended to be a prototype of a machine that will help people, who find it a problem, to climb up and down on staircases. Our project is a 30x30cm machine which adjusts a box attached to a servomotor, with high torque, while climbing up and down a staircase. The machine is completely autonomous with path tracking and obstacle detection facility that keeps the box parallel to the ground irrespective of ascend or descend of the vehicle on the staircase. The whole machine is controlled by ATMEGA 8535 which is programmed using Codevision AVR software.

Security has been a major concern nowadays. Contionuous reports of theft and roberry have been on news. Also the after theft identification of robbers too is a big problem. Hence we have made a system wherein we provide the user to access the activities going on at his home or office. Our system would also notify the end user for any unusual activity happening and also save the unusual activity on the server for the user to access.We have made the entire setup on a embedded platform. We have used a Pandaboard having TI OMAP 4460 cortex A9 multicore processor.The board is specifically used for image processing applications.The system we have made will detect any kind of motion using a USB camera and detect the face whether it is known or unknown face.If the detected face is unknown then it will call the cops or any other number provided by the user and will also save the video after motion is detected and upload it on the server with the date and time of the detected motion.The user can also access the live stream of the images if he has acsess to internet facility.The user can also access all the features using mobile phone too.

The research aims at developing an efficient handwriting recognition system used for detecting overlapped handwritten characters between adjacent lines of text. It focuses mainly on detection of the overlapped region and separating the overlapped handwritten lower case non-cursive English alphabets. Considering only the most probable cases for overlapping characters between adjacent lines of text, the recognition takes place by a novel technique where only the non-overlapped portion is considered for processing, and by reasoning which uses the knowledge of their structure, we digitize and detect the individual overlapped characters. The overlapped characters were successfully recognized with an accuracy of 85.92%. The non-overlapping characters are fed to the feed forward back propagation neural network where detection takes place. Here, diagonal based feature extraction is used.

The research aims at developing a fingerprint recognition system which is platform independent. The existing techniques uses MATLAB for its simulation. The main objective is to design a system which uses power source for its working without any PC interfaced to it, hence we use programming like C for its implementation. The data can be retrieved from the system at any desired time using the GUI designed for it. This makes the system portable and convenient for daily operations. A high level of accuracy is obtained from the designed system as the False Rejection Ratio(FRR) of the fingerprint scanner used is less than 0.1%. The proposed system is also fast as the image acquiring time is less than 0.5 secs. Moreover the system can be used as a real time security system.

A real-time system that mimics human behaviour; that is can listen or record what is spoken and then translate it to desired language has intrigued people for centuries. In order to overcome communication barrier readily available sphinx decoders are used along with already existing acoustic model tools, language model tools and creating own database with words and their translation in respective language so as to support continuous Speech to Text conversion. Further this text is given as input to phonetic synthesizer and speech synthesizer where it is matched with already existing sound database for correct speech output. It uses techniques like Hidden Markov Model and Grapheme to Phoneme rules.   Efficiency of the system is near 98% in standard pronunciation of word as input and a standard recording device.

A vision-based real-time human face sensing application is proposed for increase in safety of humans. Face Sensing needs detection and recognition for which Haar Classifiers and Scale-invariant Feature Transform (SIFT) algorithm are used respectively. Image of the person in concern is captured through a mobile phone and sent across the network to the server for detection and recognition. The input image is processed to locate the face in it using AdaBoost and then checked with each individual image in the database for the match. If the match is found then the image from the database is sent back to the requestor for double authentication. If the requestor authenticates it positively then the police is informed by reflecting the request and its details on their side. If the match is not found on one database, the image is sent to the next database in the distributed environment thus repeating the procedure until all the servers have been checked for the match. So as to avoid misuse of this application a log is maintained that keeps track of all the requests sent to the server. Efficiency of the system is near 98% in standard light conditions and a standard camera.