Implemented a VAD on a smartphone to automatically identify the background noise and update the noise reduction algorithm
Developed VAD performed better than two highly cited algorithms
VAD algorithm utilized subband features with spectrum flux, short-term energy deviation and subband power spectrum difference fed to a random forest classifier
The deployed speech processing pipeline was found to perform equivalent to manually switching the noise reduction algorithm
Apps developed for both Android and iOS mobile operating systems
Implemented a real-time noise classifier on a smartphone using subband features and random forest classifier
Deployed algorithm performed better than previously developed classifier using MFCC features with GMM classifier
Apps developed for both Android and iOS mobile operating systems
Deployed a noise-reduction app for smartphones that can be used in real-life scenarios
A post-filtering technique was applied as well to remove unwanted musical noise artifacts
Apps developed for both Android and iOS mobile operating systems
Developed a Pattern Recognition based system to classify Noise Environments using four classifiers:
-Gaussian Mixture Model
-K-Nearest Neighbors
-Artificial Neural Networks
-Support Vector Machine
The scope of the project included extensive data collection of noise samples and testing and training different classifiers based on the features extracted from the samples. The classifiers were evaluated with respect to each other.
Successfully built a hierarchical classifier based SVM followed by ANN.
Optimizations for this classifier -
-Reduced Complexity and Memory Dependency by a third
-Reduced Classification time for 1 class
Designed and developed a system to effectively generate and detect musical notes of a Piano
Identified the note by comparing the frequency, calculated by isolating the extrema by edge detection
Designed detection code in MATLAB and generation module in LabVIEW
Successfully detected single key presses of the Piano and generated an octave using a virtual keyboard