Data Acquistion Standards

[Bradley Wood Libbey <gt1556a@xxxxxxxxxxxxxxxx>]

Dear List,

For those interested in a speech acquisition discussion, I would like to
get some feedback regarding digital data acquisition standards.  My
concerns are not just superfluous, I am troubled, by reduced
intelligibility scores as a result.

I have been using an inexpensive sound card for speech acquisition and to
output speech for intelligibility tests.  (Record, process, and play using
MATLAB)  I calibrated the card "by hand" using a stand alone signal
generator and oscilloscope.  The card worked well enough as a function of
frequency and amplitude for sine waves that I felt justified in using a
single value as a calibration factor for all frequencies of the card.

However, I am concerned about noise. I suspect that a majority of the
noise actually exists as acoustic energy, but I don't want my measurement
system to be adding to this.  The first possible source of noise is from
aliased high frequency signals.  What is the standard in auditory research
for antialiasing filters?  If they are necessary, do most researchers use
stand alone filters or do they simply acquire digital data on acquisition
systems with built in filters?

Secondly is the analog circuitry or the analog to digital converter
noisier on an inexpensive card than on a "professional" acquisition
system?  In conjunction with this, the card I am using requires a low
voltage input (making the card susceptible to peak clipping), this isn't a
problem when one considers 16 bits of dynamic range is available, all I
need to do is turn down the gain stage on the preamplifier.  However, some
of this range is used up by the internal circuitry noise floor.  I'm
either cutting the peaks or burying the quiet in noise.  In other words
limiting my dynamic range.

Any comments, discussion, or references on digital equipment standards for
speech testing would be useful.

Sincerely,
Brad Libbey
Georgia Institute of Technology



On Thu, 14 Jun 2001, Bob Masta wrote:

> On 14 Jun 01, at 2:51, Tom Brennan wrote:
>
> > Of course, the problem with using a sound card for auditory output is that it is
> > not calibrated and even if it was there would be no way to calibrate individual
> > speakers other than manually and no way to test for pure tones and output
> > linearity.
> >
>
> Of course, *any* system needs to be calibrated.  You would certainly want to
> calibrate any speaker or headphone used for research purposes, whether
> "manually" or with some automated system.  In this respect a sound card
> fares no worse than an expensive lab-grade acquisition system.  As for pure
> tones and output linearity, my Daqarta software does real-time spectral
> analysis, so it's pretty easy to check these.
>
> And unlike many lab-type boards, a Sound Blaster has a built-in attenuator.
> In fact, it is better than many older lab-type manual attenuators as far as
> accuracy goes, since it doesn't suffer from leakage/ground-loop problems that
> often arise with rack-mounted units.  The steps are very accurate... it's only
> Windows that keeps you from knowing about that!
>
> By the way, the next version of Daqarta (v2.10, out in late July) will include
> enhanced facilities for automatic calibration, including in-ear calibrations and
> probe tube transfer functions.  It's also specifically designed to make it
> simple to do a transfer calibration from a reference mic to a "working" mic.
>
>
> Robert Masta
> tech@daqarta.com
>
>          D A Q A R T A
> Data AcQuisition And Real-Time Analysis
>  Shareware from Interstellar Research
>         www.daqarta.com
>