Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Physical Structures for Signal Separation

Nordsjö, Albin LU (2016) EITM01 20161
Department of Electrical and Information Technology
Abstract
This master’s thesis studies the use of physical structures for signal source separation. A parabolic reflector was used to alter the directional properties of one microphone in a two-microphone array. A method to estimate the mixing channels from measured data is presented, as well as a method to find the optimal separating channels. Measurements of the directional properties and mixing channels in a two-source, two-sensor set-up were made using three reflectors of diameters 10, 5 and 2.5 cm. Speech and noise was mixed with the estimated channels and then un-mixed with the optimal separating channels. The performance of the separation was objectively evaluated, and the results were used to determine the impact of using reflective... (More)
This master’s thesis studies the use of physical structures for signal source separation. A parabolic reflector was used to alter the directional properties of one microphone in a two-microphone array. A method to estimate the mixing channels from measured data is presented, as well as a method to find the optimal separating channels. Measurements of the directional properties and mixing channels in a two-source, two-sensor set-up were made using three reflectors of diameters 10, 5 and 2.5 cm. Speech and noise was mixed with the estimated channels and then un-mixed with the optimal separating channels. The performance of the separation was objectively evaluated, and the results were used to determine the impact of using reflective structures. The results showed that the 2.5 and 5 cm reflectors had very little effect for sound in the voice frequency range, but the 10 cm reflector improved the separation to a certain extent. Thus, there is a potential in using physical structures for signal separation purposes, however the frequency of the sound puts a limit on how small they can be made. (Less)
Popular Abstract (Swedish)
Att separera två eller flera inspelade ljudsignaler är generellt ett svårlöst problem, men det blir enklare att lösa desto mer signalerna skiljer sig från varandra. Genom att använda en parabolformad reflektor går det att
skapa skillnader mellan signalerna på ett sådant sätt att signalseparationen kan förbättras.
Please use this url to cite or link to this publication:
author
Nordsjö, Albin LU
supervisor
organization
course
EITM01 20161
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Signal separation, physical, parabolic reflector, reflector, signal processing
report number
LU/LHT-EIT 2016-527
language
English
id
8885139
date added to LUP
2016-06-28 10:01:35
date last changed
2016-06-29 15:59:17
@misc{8885139,
  abstract     = {{This master’s thesis studies the use of physical structures for signal source separation. A parabolic reflector was used to alter the directional properties of one microphone in a two-microphone array. A method to estimate the mixing channels from measured data is presented, as well as a method to find the optimal separating channels. Measurements of the directional properties and mixing channels in a two-source, two-sensor set-up were made using three reflectors of diameters 10, 5 and 2.5 cm. Speech and noise was mixed with the estimated channels and then un-mixed with the optimal separating channels. The performance of the separation was objectively evaluated, and the results were used to determine the impact of using reflective structures. The results showed that the 2.5 and 5 cm reflectors had very little effect for sound in the voice frequency range, but the 10 cm reflector improved the separation to a certain extent. Thus, there is a potential in using physical structures for signal separation purposes, however the frequency of the sound puts a limit on how small they can be made.}},
  author       = {{Nordsjö, Albin}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Physical Structures for Signal Separation}},
  year         = {{2016}},
}