CONTRACT RD-94 TASK ORDER NO. 2

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP78-03300A001900110010-7
• Release Decision: RIPPUB
• Original Classification: S
• Document Page Count: 5
• Document Creation Date: December 22, 2016
• Document Release Date: May 3, 2011
• Sequence Number: 10
• Publication Date: August 16, 1957
• Content Type: LETTER

Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
August 16, 1957
25X1
25X1
SubJoets Contract RD-94
Task Order No. 2
In aocordas>ou with Article 2 of the basic contrast,
there are forwarded herewith two (2) copies of the l thly
Progress Report for July, 1957 on Task Order No. 2 of ID-9.
This r I. UICLAS8IP3 D. An additional copy is being
held in by the pro jest a*inser for the use of 25X1
7oor porsonne while at this location.
In ea ostion with this nenthly propose report,
the following intomation is subuittods
Total expenditures to 6-30-57 $37A9
Outstanding coaoaitment as
6-30-57 65
Panda remaining as of 6-30-57 $22,602
You will note that the figures given above are the
sasses as the" reported for 5-31-57. The figures shown for
5-31-57 in our letter of July 17, 1957, should have reads
Total expenditures to 5-31-57 633,413
Outstanding aomttwsnts as of
5-31-57 lone
Ponds remaining as of 5-31-57 126,903
very truly yours,
ILLEGIB 25X1
Manager
Oovormseat Contract Administration
TR2 snit
f-2.60
Enclosures
~,~ , ~Vo_~
Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
 Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
LU1 I"JVtIVI IAL
Monthly Progress Report
July 1957
Task Order No. 2
Contract No. RD-94
Audio Noise Reduction Circuits
The object of this project is to develop a noise reduction circuit
suitable for use in separating speech intelligence from a signal containing
speech and noise when the speech intelligence is masked by the noise. The
proposed method involves a principle which has been used successfully to,
improve the signal-to-noise ratio in music ~eeproduoing or transmission
systems.1 The system used for music contains bandpass filters which pass
frequencies over a range of an octave or less. These filters are used
at the input and output of a none-linear eler,;,ent. The output of the non-
linear elements contain the fundamental, and also harmonics and subharmonios
of the fundane:atal. however, since the past: band of the input and output
bandpass filters is no greater than an oota'e, the harmonics and subharmonics
are not transmitted by the system. Iles funetion of the non-linear element
is to reject all noise signals below a given amplitude or threshold level.
The threshold :Levels of the non-linear devices in each channel can be
>bdjusted so that, in the absence of desired signal, the noise is rejected.
;then the desirc,d signal is greater than the threshold level, the non-linear
~olerr.nts allow the composite s.gnal to pass. Thus, for passages of recorded
rusl.c,,, when they music signal its below the noise level in a given frequency
,3hanncl, the channel is inoperative, and its output is eliminated from the
total output. Since -dm contribution of this channel to the total output
would have been only noise, the over-all noi!:,e level is reduced. When the
H.F, Olson, "Electronics," Dec. 1947.
CONFIDENTIAL
Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
 Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
munio signal in a given channel is greater than the noise. the channel
conducts and allows the composite signal to pass. Thus, a channel conducts
only when the desired signal is greater than the noise,, and rejects when
no'kae alone is present.
In order to apply th:.s method of r!oise reduction to speech, when
the -wide band speech signal-to-noise ratic is very low, it was believed
necessary to find frequency regions in which the speech amplitude is
granter than the noise. Although the long time average spectrum of speech
is continuous, and similar in shape to the spectrum of room rroiae,2 the
short time spectrum of various speech sounds contains regions of maxim-
energy called speech formants.3 The assumption that this method of noise
reduction should be utilized for speech was based upon the belief that it
would be possible to find frequency regions in which the amplitude of the
speech Torments would be greater than the noise a substantial part of the
A study has been made to determine what bandwidths are required in
order to obtain speech formant amplitudes above the noise when a wide band
speech sample is just intelligible in noise. It is known that for noises with
a continuous spectrum it is the noise components in the immediate frequency
region of the masked tone which contribute to the masking.4 When a very
narrow band of noise is used to mask a pure tone, the masking increases as
the bandwidth is increased until a certain bandwidth is reached. After this,
as the bandwidth is increased, the amount of masking remains constant. This
bandwidth at which the masking reaches a fixed value is termed the critical
2. U. Fletcher, "Speech and Hearing on Cosamunioation," Van Noutrand Co., Inc.,
Now York, 1953 (see Figures 61 and 70),
3. Op.oit. chap. 1.
4. I..L. Beranek, "The Design of Speech Car~,muuication Systemn," pron. IRE.
Vol. 35, pp. 882. Sept. 1947.
Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
 Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
UUIVrIL)U IIIL
bandwidths Measurements have been made using filters which were both
narrower and wider than the critical bandwidth. Both pure tones and speech
mixed with continuous spectrum type noises have been studied. The results
of this study show that, for the narrowest permissible bands which can be
used to pass speech fbrmants, the number of times the speech formant
amplitude in a given band exceeds the noise is small. Also, in these bands,
the speech amplitude is never considerably greater than the noise. Since
very narrow bandwidths are required to reduce the noise below the signal,
the number of bands required to cover the speech spectrum is quite large.
There was no satisfactory way of evaluating the effect upon speech intelligence
of small contributions from many narrow bands without building a many
channeled circuit and evaluating it.
In view of this fact a complete multi.ohannel system has been
developed in order to determine the effectiveness of this method of improving
speech intelligibility in noise. The multi-channel system developed contains
80 channels covering the frequency range from 700 to 3200 ops. The bandwidth
of each channel is adjustable and each has been set so that it is one half
that of the critical band when the signal is 3 db above the threshold level.
A schedule of the band centers and their bandwidths is contained in the
progress report for January 195?. A preliminary evaluation of the circuit
has been performed.
During July evaluation of the noise reduction circuit was continued.
A series of tape recordings were made of the noise reducer output for various
test conditions. These recordings will provide a permanent record of the
effectiveness of the noise reduction circuit. Tests were made for signals
5. N.R. French and J.C. Steinberg, "Factors Governing the Intelligibility
of Speech Sounds," Jour. Aooust. Soo. Amer., Vol. 19, Jan. 1961
(See Figure 7).
O P, I
F11FIDENTIAL
Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
 Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7
CONFIDENTIAL
consisting of speech in noise at various levels above the threshold in
5 db steps. Both flat or white noise and simulated room noise were mixed
with speech so that the speech was judged to be just intelligible in the noise
before it was passed through the noise reduction circuit. The thresholds
of the noise reducer were set either flat, i.e., all equal] or varied as
a function of frequency so that they followed the shape of the room noise
spectrum. The results of these tests indicate that the signal to noise
ratio of the noise reducer output. as viewed on a scope is greatly improved,
but the improvement of the intelligibility of speech is marginal.
It has previously been found that the peak amplitude of a speech
signal at be about 15 to 20 db above the threshold of the noise reduoer.
Originally the bandwidths of all channels were adjusted for the desired
bandwidth when the signal was 3 db above threshold. The bandwidths of the
channels are a function of amplitude so it is conceivable that a narrower
bandwidth would be more effective for signals 20 db above threshold. In
view of this the bandwidths of all channels have been made narrower. Tests
for this condition have not yet been made.
During July thirty additional channels were delivered by the model
shop. These channels have been tuned and adjusted and are now ready to be
added to the low frequency end of the noise reducer. With the addition
of these ohannels the noise reducer will have 110 channels covering the
frequency range from 170 to 3200 ops. Evaluation of the complete system
is being continued.
August 7, 1957
CONFIDENTIAL
Sanitized Copy Approved for Release 2011/05/03: CIA-RDP78-0330OA001900110010-7