Estimation of NonGaussian Acoustic Channel for Array of Microphones
This project aims to estimate the different noise models, Gaussian and Non-Gaussian models (Gaussian Model, Alpha-Stable Model, Mix-Gaussian Model) in different Scenarios (Auditorium, Outdoor, Hall) and compare them. the collection of data audios to analysis were realized in the Federal University of Rio Grande do Norte (UFRN) at the NPITI lab,
these are the results plots of each scenario, the histogram, Power Spectral Density (PSD), time and frequency domain of the first channel datas and of each kind of estimation.
AUDITORIUM NPITI
Environment
This environment is the NPITI’s auditorium which is a indoor with soundproof. It’s a place where the UFRN’s teachers and students can lecture, So it’s a normal and quite place without large noises.
Auditorium NPITI  |
Data Analysis
Histogram  |
PSD  |
Time and Frequency Domain  |
Normal Estimation
Normal distribution parameters:
mu=0.0000446289
sigma=0.0039705515
Histogram Gaussian Estimation  |
PSD Gaussian Estimation  |
Alpha-Stable Estimation
Alpha stable parameters:
alpha=1.9872
beta=0.3390
gamma=0.0027
delta=0.0000
PSD Auditorium Alhpa-Stable  |
PDF Auditorium Alhpa-Stable  |
Histogram Auditorium Alhpa-Stable  |
Gaussian Mixture Estimation
Parameters of the Gaussians:
Mu1: -0.0011067
Sigma1: 0.0036058
Mu2: 0.0013198
Sigma2: 0.0035814
Gaussians of Mix-Gaussian Auditorium:  |
Gaussians Sum of Mix-Gaussian Auditorium:  |
we can see that for quite places the normal distribution is better.
OUTDOOR NPITI
Environment
This environment is outside of the NPITI building, so it’s an outdoor environment and also a parking lot. There was construction next to the collection data equipment (Respeaker-Raspberry). Therefore it’s a loud place with significant noises.
Outdoor NPITI  |
Outdoor NPITI  |
Data Analysis
Histogram  |
PSD  |
Time and Frequency Domain  |
Normal Estimation
Normal distribution parameters:
mu=3.792169e-05
sigma=4.744719e-02
Histogram Gaussian Estimation  |
PSD Gaussian Estimation  |
Alpha-Stable Estimation
*building open-field environment
Alpha stable parameters:
alpha=1.2670
beta=0.0051
gamma=0.0167
delta=0.0002
PSD Outdoor Alhpa-Stable  |
PDF Outdoor Alhpa-Stable  |
Histogram Outdoor Alhpa-Stable  |
Gaussian Mixture Estimation
Gaussians of Mix-Gaussian Outdoor:  |
Gaussians Sum of Mix-Gaussian Outdoor:  |
we can see that for laud places the normal distribution is worse and the alpha-stable is better.
HALL NPITI
Environment
This environment is inside of the NPITI building and the collection of datas had been made by the night, so there aren’t so much noise, but there are reverberation because of the halls.
Hall NPITI  |
Hall NPITI  |
Data Analysis
Histogram  |
PSD  |
Time and Frequency Domain  |
Conclusion
This project was about the characterization of the noise and how it works in every signal to have the better characterization for any situation. Every environment is specific and needs to have a particular representation of the noise in every single case.