MASLA Microphones Arrays for Source Location Applications

1
MASLAMicrophones Arrays for Source Location
Applications

• 18-551
• Group 1, Final Report
• Akhil Chauhan (achauhan_at_)
• Bryan Chen (bryanche_at_)
• Anup Doshi (adoshi_at_)
• Vinay Kapur (vinayk_at_)
• April 26, 2004

2
Overview of Project

• Using the concept of Microphone Arrays
• Detect the speakers angle from two different mic
  setups.
• Using this pair of angles, detect the speakers
  distance from the designated origin mic
• Thus, we can successfully determine the speakers
  location

3
Numerous Applications

• Accurately identifying location based solely on
  speech has numerous applications
• Security (Intruder Detection)
• Military Applications
• Wildlife Research Tracking animals, birds etc
• Automating camera direction in conference calls
  to point at the speaker

4
Our Story

• Identifying speaker location for the purpose of
  videoconferencing
• Four mics positioned on a conference table
• As people speak, a graphical display will
  continuously output the position of the speaker

5
Prior Work

• No Prior 551 project has used microphone arrays
  for source location.
• Several papers have been written on this topic.
• M. Omologo and P. Svazier. Use of
  Crosspower-Spectrum Phase in Acoustic Event
  Location. IEEE Transactions on Speech and Audio
  Processing, May 1997
• Rabinkin, Renomeron, Dahl, French, Flanagan and
  Bianchi. A DSP Implementation of Source Location
  Using Microphone Arrays. Proceedings of the
  SPIE, Denver, August 1996

6
Our Solution

• Using 4 mic inputs and the Cross-Power Spectrum
  Phase Spectrum we obtain the TDOAs
• Using the TDOA for each pair of mics we calculate
  the angle from the sound source to the mic
• Using the two angles we also calculate the
  distance.

7
Cross-Power Spectrum Phase

• CSP method is the method being adopted in this
  project to calculate TDOAs.
• We could just correlate the two signals received
  at the two mics, but we get better results using
  CSP.
• WHY?
• We require only the phase information.
• CSP gives a better peak to side lobe ratio.
• CSP gives a better signal to noise ratio(SNR).
• Signal to noise ratio is the ratio between
  the highest peak and the average of all other
  peaks.

8
CSP cont.

• We divide by magnitude of fft because we are only
  concerned about the phase.
• We perform this calculation for two different
  pairs of mics to get two different time
  delays(One per each doublet).
• Using the two different time delays we get the
  angle indicating the wavefront arrival direction
  from each pair.

9
Signal Flow

• From speaker to 4 mics
• 2 mics are direct to EVM thru Mic In
• 2 mics go into a USB port via an adapter we
  purchased, PCI transfer places these on the EVM
• Run Algorithm on Input.
• Results will go out via PCI Transfer
• Visual C program will indicate speaker location
  from origin microphone.

10
Signal Flow cont.

• C is used to record 0.5 seconds of sound into a
  buffer from the mics that are connected into the
  soundcard.
• The Buffer is then transferred into the EVM.
• Meanwhile, The EVM records 0.5 secs of the sound
  and does the CSP and angle calculations for both
  pairs of mics.
• Once both angles are calculated on the EVM, they
  are transmitted to the PC.

11
Signal Flow 4 Mic Challenge
12
Sampling Rates

• Sample .5 seconds of speech on 4 mics at 8 KHz
• Upsampling requires too much memory / is too
  slow.
• Quick Calculations
• Buffer size 4096 samples 8 bytes/sample
  (complex floats) 2 channels 64kb memory
  required per pair
• PCI Transfers of 16kb each time from PC to EVM
  (short ints).
• CPSP code cycle count 44080696 (_at_133 MHz .3
  sec)

13
Memory Issues

• FFT requires buffers be in complex float format.
• Algorithm does not work well for small buffers.
• So we came to a compromise of .5 seconds (4096
  samples)
• All buffers do not fit on ONCHIP memory.
• We had to work around this by placing things on
  SBSRAM and SDRAM0.
• This led to a compromise in speed.

14
Memory/Speed Issues

• Since we record 0.5 seconds of sound the EVM has
  a chance to catch up on calculations.
• There is a lot of on-board memory transfers
  required since much of the data lies in SDRAM0.
• However this does not have a significant effect
  on the speed of the program.
• Optimization has adverse effects on
  Synchronization/Timing

15
Synchronization

• Control of the FFT function was split between the
  interrupts (corresponding to mics going into the
  EVM) and PCI transfer callback function
  (corresponding to the mics going into the PC).
• We had to put some delays in the code to maintain
  synchronization.
• That causes a delay in the output.

16
Distance Calculation/Sampling Rate Problem
17
(No Transcript)
18
Lab Demo

• Real time updates of the position of sound source.

19
Purchases

• Stereo Mics required for the left and right
  inputs, allows us to have 4 separate mics into
  two inputs
• iMic serves as a preamp, as the mics were not
  working directly into the existing soundcard

20
References

• M. Omologo and P. Svazier. Use of
  Crosspower-Spectrum Phase in Acoustic Event
  Location. IEEE Transactions on Speech and Audio
  Processing, May 1997
• Used to obtain and understand most of the
  algorithm
• C Code
• http//www.codeproject.com/audio/fister.asp?target
  play7C2Brecord7Csound
• How to play and record soundBy Thomas Holme
• M. Brandstein and D. Ward. Microphone Arrays,
  Springer 2001
• Basic concepts and issues that could arise (i.e.
  reverberations)