Ranjan Acharyya Email & Phone Number

New Zealand

Identify appropriate vendor to carry out air monitoring of GIQ buildings, liaise with building occupants, set up air monitoring programme,,provide reports and update ELT on status of air quality, develop improvement action plan as required.

· The principle of Rough path theory has been applied into the core signal processing area. The novel algorithm is developed to detect a target which would otherwise be hidden under a sidelobe. The concepts from Rough Path theory can be borrowed to improve SNR for multidimensional system. This link shows an example: https://dsp.stackexchange.com/questions/93291/is-this-better-detection-than-matched-filter-and-gaussian-noise-cancellation-tecMachine Learning Paradigms: Unconstrained vs. Constrained Resource Environments:· Unconstrained Resources:· Developed and tested a speaker recognition model using Keras and TensorFlow, scikit-learn etc· Extracted Mel-Frequency Cepstral Coefficients (MFCC) and I-vectors as features for speaker identification.· Conducted a comparative analysis of MFCC and I-vector features to evaluate their performance in the recognition task· Constrained Resources (Node-Level)· Developed a Binary Neural Network (BNN) for speaker recognition: Implemented a BNN to perform efficient and accurate speaker identification using binary weights and activations, optimizing the model for reduced memory usage and faster computation.· Integrated BNN with Edge Devices: Successfully deployed the BNN-based speaker recognition system on edge devices and mobile platforms, ensuring efficient operation and real-time processing in practical applications.· Implemented Sequential Minimal Optimization (SMO) and compared its performance with support vector machine model using convex programming CVX. Applied the learning algorithm for automatic detection of queen in the beehive.

· Implemented a time frequency masking algorithm for extraction of fetal heartbeat using noninvasive technique. The algorithm uses the sparseness of data in the time-frequency plane. The separation of the desired signal was also performed using Independent Component analysis based on maximization of the higher order statistics.· Designed a filter to clean signals collected from devices to measure breathing rate. The sensing device is around a strap that is worn on the chest. The device was worn while sleeping only.

· Developed a classifier using parallel Radial Basis Function (RBF) Neural Network bank for classification of man-made signals linked to military activities using Infrasound data. A comparison of classification performance among Support Vector Machine (SVM), HMM and RBF is made for Infrasound data.· Enhanced the performance of the RBF based classifier using statistical analysis to obtain optimal network parameters.· Introduced a novel feature selection method for RBF neural network. Attempts have been made in the selection process so that the input elements may become a member of RKHS. The new technique improved the performance measures of the classifier considerably.

· Developed mathematical model to predict the resonance of structures subjected to longitudinal infrasound waves. Utilized MATLAB for implementation. Finite element software, FEMLAB, is used for validation.· Studied the resonance behavior of pipe under different terminating conditions. Appropriate mathematical forms of the Boundary conditions were developed based on their physical characteristics.· Found a unique method to find radius and length of a pipe lying on ground based on the observed radiated signal.· Developed a model in COMSOL to simulate far field characteristics due to a generator running inside a room. Multiphysics mode of COMSOL was utilized.

· Implemented Independent Component Analysis (I.C.A.) algorithm in wavelet domain. The algorithm can separate instantaneous as well as artificially convolved speech sources. Information maximization principle and feedback network architecture is utilized. Adaptive weight updating is based on maximum information flow through the network.· Developed a novel, post processing algorithm for separation of convolved sources. Tested with sources composed of speech and music signals recorded inside a room under reverberant condition. The algorithm reduces the cross talk, which is an inherent problem in BSS. The cross-talk component is reduced using GMM along with a shrinkage function.

· Designed a voice activity detector (VAD) based on MFCC. Implemented the algorithm in C and MATLAB.· Developed speech recognition algorithm using MFCC and DTW for real time applications· Implemented an algorithm for generating Mel Frequency Cepstral Coefficient (MFCC) from continuous speech signals in MATLAB and C.

Ph.D, Electrical Engineering

Master Of Science, Electrical Engineering

Bachelor Of Science, Electrical Engineering

Education record