Bill Cole: Speech-shaped noise

[Dan Ellis <dpwe@xxxxxxxxxxxxxxx>]

Dear List -

I am forwarding a response from Bill Cole of Etymonic/Audioscan
concerning issues in generating "speech-shaped noise".

  DAn.

------- Forwarded Message
From: Bill Cole <bill@xxxxxxxxxxxx>
Subject: Speech-shaped noise
Date: Sun, 20 Feb 2005 14:00:32 -0500

Dear Richard:

Speech-shaped noise will not do you much good since it is unmodulated.
You also need to be specific about the frequency shaping that is applied
to the noise since the ANSI S3.42 shape is intended to represent the
spectrum of speech peaks whereas the shapes in ANSI S3.5 (and used for
the ICRA signal) is representative of the LTASS of speech for various
vocal efforts.  If you have the facilities, I would suggest taking a
short (10 - 15 s) passage from a recording of everyday speech and
filtering it to produce the LTASS you want.  Alternately, you might
consider the ICRA digitally-modified speech signal
(www.icra.nu/ITEMS/ICRAnoise.htm ). It has some shortcomings but it
gives results similar to real speech for any aids I have tested

It is not just the signal that matters but also the way in which it is
analyzed.  If you really want to know what is going on with digital
hearing aids, you need to use real speech as the test signal with
analysis parameters that permit comparison with behavioral measures of
hearing.  If you have the capability to perform 1/3 octave spectral
analysis, I would suggest setting your analysis parameters to yield 128
or 256ms analysis blocks to approximate the integration time of the ear.

We sell a product (Verifit) which accomplishes these goals.  You can
read about it at www.audioscan.com and in the October and November
issues of the Hearing Review at www.hearingreview.com .  If you care to
contact me directly, I can provide some additional technical details of
the signals we use and their analysis.

Regards,

Bill Cole


------- End of Forwarded Message