Advanced

Increasing Sound Quality using Digital Signal Processing in a Surveillance System

Johnsson Henningsson, Gustav LU and Siljeholm, Patrik (2016) EITM01 20161
Department of Electrical and Information Technology
Abstract
Hearing is our second most important sense. The main focus of a surveillance camera is obviously the video but the audio shouldn't be neglected. It is possible to hear what the camera can't see and what is said in a conversation can't be showed in a video. Therefore there are good reasons to always try and achieve the best audio possible. This is not an easy task since cameras are operating in a plethora of different environments with different background noises and are mounted with varying distances to the origin of the sound.


This thesis investigates which possibilities there are for improving the audio quality using digital signal processing. The thesis consist of three main parts: the choice of microphone, equalization and noise... (More)
Hearing is our second most important sense. The main focus of a surveillance camera is obviously the video but the audio shouldn't be neglected. It is possible to hear what the camera can't see and what is said in a conversation can't be showed in a video. Therefore there are good reasons to always try and achieve the best audio possible. This is not an easy task since cameras are operating in a plethora of different environments with different background noises and are mounted with varying distances to the origin of the sound.


This thesis investigates which possibilities there are for improving the audio quality using digital signal processing. The thesis consist of three main parts: the choice of microphone, equalization and noise reduction. The microphone is the first link in the audio system and influences the conditions for digital signal processing. An equalizer was implemented to compensate for the acoustics in and around the microphone to make the audio sound more natural and closer to the original sound. The equalizer estimates the frequency response from the microphone in the frequency domain and inverts it in order to get a flat frequency response. Surveillance cameras are stationary and not seldom there is some kind of source of stationary noise nearby, e.g. ventilation or power switching from the camera itself. From the user's point of view it can be tiresome and hard to distinguish different sounds, why noise reduction is desired. Four different noise reduction algorithms were investigated and implemented and finally one algorithm was chosen that produced the best result, which was Multi-band Spectral Subtraction. Since all non-stationary sounds should be left untouched and only stationary noises removed the noise is estimated over a long time. Different ways of reducing the musical noise which often occurs after noise reduction are tested and the listening experience was in focus when tweaking the parameters. Everything is implemented in real-time Matlab. (Less)
Please use this url to cite or link to this publication:
author
Johnsson Henningsson, Gustav LU and Siljeholm, Patrik
supervisor
organization
course
EITM01 20161
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Digital Signal Processing Noise reduction Equalizer Microphone Spectral subtraction Surveillance Audio Wiener filter
report number
LU/LHT-EIT 2016-516
language
English
id
8879354
date added to LUP
2016-06-21 08:17:40
date last changed
2016-06-21 14:42:46
@misc{8879354,
  abstract     = {Hearing is our second most important sense. The main focus of a surveillance camera is obviously the video but the audio shouldn't be neglected. It is possible to hear what the camera can't see and what is said in a conversation can't be showed in a video. Therefore there are good reasons to always try and achieve the best audio possible. This is not an easy task since cameras are operating in a plethora of different environments with different background noises and are mounted with varying distances to the origin of the sound.


This thesis investigates which possibilities there are for improving the audio quality using digital signal processing. The thesis consist of three main parts: the choice of microphone, equalization and noise reduction. The microphone is the first link in the audio system and influences the conditions for digital signal processing. An equalizer was implemented to compensate for the acoustics in and around the microphone to make the audio sound more natural and closer to the original sound. The equalizer estimates the frequency response from the microphone in the frequency domain and inverts it in order to get a flat frequency response. Surveillance cameras are stationary and not seldom there is some kind of source of stationary noise nearby, e.g. ventilation or power switching from the camera itself. From the user's point of view it can be tiresome and hard to distinguish different sounds, why noise reduction is desired. Four different noise reduction algorithms were investigated and implemented and finally one algorithm was chosen that produced the best result, which was Multi-band Spectral Subtraction. Since all non-stationary sounds should be left untouched and only stationary noises removed the noise is estimated over a long time. Different ways of reducing the musical noise which often occurs after noise reduction are tested and the listening experience was in focus when tweaking the parameters. Everything is implemented in real-time Matlab.},
  author       = {Johnsson Henningsson, Gustav and Siljeholm, Patrik},
  keyword      = {Digital Signal Processing Noise reduction Equalizer Microphone Spectral subtraction Surveillance Audio Wiener filter},
  language     = {eng},
  note         = {Student Paper},
  title        = {Increasing Sound Quality using Digital Signal Processing in a Surveillance System},
  year         = {2016},
}