Scribd
Upload
94 views3 pages

Immersive Audio Signal Processing

This document discusses live multi-track audio recording and a method to resolve individual sound sources from each microphone recording. It aims to separate out the "dry sound" from each source, which gets mixed together due to each microphone picking up sound from multiple sources simultaneously during a live recording. The method models the recording system using linear algebra equations to represent the mixing of different sound sources based on room impulse responses measured between each source and microphone. Experimental results show the method successfully removes cross-talk between channels compared to simply equalizing each channel individually. More accurate room response measurements could improve the signal-to-noise ratio of the recovered dry sounds.

Uploaded by

Subha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
94 views3 pages

Immersive Audio Signal Processing

This document discusses live multi-track audio recording and a method to resolve individual sound sources from each microphone recording. It aims to separate out the "dry sound" from each source, which gets mixed together due to each microphone picking up sound from multiple sources simultaneously during a live recording. The method models the recording system using linear algebra equations to represent the mixing of different sound sources based on room impulse responses measured between each source and microphone. Experimental results show the method successfully removes cross-talk between channels compared to simply equalizing each channel individually. More accurate room response measurements could improve the signal-to-noise ratio of the recovered dry sounds.

Uploaded by

Subha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Live recording of multi-track

audio

Motivation
Studio recording
Each sound track is recorded separately

Joao Luiz Carvalho

Live Recording
Concerts, band practice
All tracks recorded simultaneously
Each microphone perceive sound from
multiple sources
Overall quality is lower

EE522 Project

04/24/2007

Objective

System Model
H1A

Resolve the dry sound coming from each


source in a live recording

H2B

H3C
H4D

yA= x1 * h1A + x2 * h2A + x3 * h3A + x4 * h4A

System Model
x1

H1A

H1A

ya

2A

3A

x3

H3C

x2

H 4A

3A

H2B
H 4A

2A

x4

H4D

System Equations

Frequency Domain Linear System

yA= x1 * h1A + x2 * h2A + x3 * h3A + x4 * h4A

YA= X1 H1A + X2 H2A + X3 H3A + X4 H4A

yB= x1 * h1B + x2 * h2B + x3 * h3B + x4 * h4B

YB= X1 H1B + X2 H2B + X3 H3B + X4 H4B

yC= x1 * h1C + x2 * h2C + x3 * h3C + x4 * h4C


yD= x1 * h1D + x2 * h2D + x3 * h3D + x4 * h4D

YB
YC

YD

YD= X1 H1D + X2 H2D + X3 H3D + X4 H4D

System solution

Matrix representation
YA

YC= X1 H1C + X2 H2C + X3 H3C + X4 H4C

Solve for each frequency component:

H1A H2A H3A H4A

X1

H1B H2B H3B H4B

X2

H1C H2C H3C H4C

X3

X1

H1A H2A H3A H4A

X4

X2

H1B H2B H3B H4B

YB

H1C H2C H3C H4C

YC

H1D H2D H3D H4D

YD

H1D H2D H3D H4D

X3
microphone
recordings

logsweep
measurements

dry sounds
(unkowns)

Experiments - Limitations

X4

-1

YA

Experimental Setup

Lots of noise
Repeated each recording 10 times

Mic+Speaker System
Low and high frequencies very attenuated
Inverted response: very high gains (100dB +)
Noise amplification
Focused on 100 Hz to 8kHz band only

Could only record one track at a time


Had to improvise

Recordings

Results

Used Adobe Audition to obtain


synchronized recordings
16 logsweeps (hij)
16 audio clips (yij)
yA = y1A + y2A + y3A + y4A
yB = y1B + y2B + y3B + y4B
yC = y1C + y2C + y3C + y4C
yD = y1D + y2D + y3D + y4D

Simulations showed that this can


be improved with more accurate
room response measurements

Auto-correlations
Cross-correlation between the 4 audio
clips (maximum value)
dry vs. dry

dry vs. equalized

SNR (dB)
Eq Inv
Ch1 9.4 12.0
Ch2 6.4 11.2
Ch3 -0.4 11.6
Ch4 8.5 12.0

dry vs. recovered

Conclusions
Proposed method successfully removed
cross-talk
Consistently better than simply equalizing
each channel
SNR can be improved using more
accurate room response measurements
Thank you!

Cross-talk successfully removed


Joao Carvalho 2007
jcarvalh@usc.edu

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy