Aarsh Prakash Agarwal 150004 [email protected] Harshvardhan 150283 [email protected] Shubhanshu Khandelwal 150705 [email protected] Krishna Chandra 150337 [email protected] Vishal Srivastava 150820 [email protected]
This is the main script for the second paper's implementation. It defines the 3 parts of the algorithm, source motion, sampling possible source positions and updating the weights for the sampled source position using SSP_EM function on the convolution of the RIR and STFT of the observed speech signal. Then it plots the KDE contour plots for each time step.
Function takes in the time domain signal values, sampling rate, window, and hop length(here it is the same as window length) and outputs the STFT of the given signal. This function has been ported from MATLAB File Exchange.
Function has been ported from the RIR library. It computes the RIR for given source-microphone geometries when the reverberation time and room dimensions have been supplied.
This file runs the subroutine EM algorithm for at max 50 iterations. It uses certain threshold for the convergence. This file assumes following input from file sampler.m
1. microphone position
2. sampled state(x,y,v_x,v_y) of particles
3. STFT of the signal recieved by microphones.(size = 2 X K; K = no. of frequency bins.)
4.
Function that takes point and covariance matrix (as a complex number) as input and returns the probability of the point from the circular Gaussian distribution with zero mean and given covariance matrix.
Function that takes two matrices as input and returns the maximum of the absolute value of the difference between the two.
There are basically two files. The first file implements the algorithm and plots the spectrum while comparing its with state of the art MUSIC and MVDR algorithms, while the second file implements the Monte Carlo simulation and compares their performance on different SNR values. Following is the description:
The file implements the signal model and IG algorithm, and compares its spatial spectrum with MUSIC and MVDR. MUSIC and MVDR are implemented via in-built phased library functions present in MATLAB.
The file implements the Monte Carlo simulation for all the three algorithms. The simulation is run for different values of SNR depending upon the problem. Sources are kept to 2 in this file.