W6OO 7()3 - DTIC

[Pages:18]AD-A011 709 STUDY AND DEVELOPMENT OF THE INTEL TECHNIQUE FOR IMPROVING SPEECH INTELLIGIBILITY M. R. Weiss, et al

Nicolet Scientific Corporation

Prepared for: Rome Air Development Center

April 1975

Reproduced From Best Available Copy

`0

DISTRIBUTED BY:

National Technical Information Service U. S. DEPARTrM!EJllT OF C?'1,i!1,[RCE

W7(6)3OO

,1

1 95061

STUDY MMD VELCUP?T OF THE IHTEL TECIMIQUIE FOR IM. OVIMG SPEECH INTELLIGIBILITY licolet Scientific Corporation

Approved for public release; d/etribution ual/lued.

Laboratory Directors' Fund No. 01737401 Roam Air Deelopment Center Air Force Systems Comand

Griffis& Air Force Uise, Nev York 13441 NATIONAL TECHNICAL INFORMATIO.4 SERVICE

$p.g~VA 2215

-

7

T41m rc,,-v-t hma b,?-n rzvfmcd by tl,4 IUJC Info trtic'a Office (01)

and

reIc.la bla to the rtlatioenr Teehnical Thfo t

Sctvicc (-17 ).

At r71 iz wvil be rehuzble to tb-i gncrrl

public, ikncludiris forcign

This report haut bcn revicwed and approved for publication.

Ro-I0,L A. CZTTIS, Capt, USAF Project Eqineer

hPPROVtD: HOWARJ) DAVIS Technical Director Intel .& Recon Division

]FOR TUB aOJJIDER:

S;

JO? P HUSS? Acting Chief, Plans Office

This effort was funded totally by the Laboratory Directors' Fund. Do not return this copy; retain or destroy.

ae

'I

UNCLASSIFIED

SECURITY CLAWSIFCATILN OF THIS PAGE (W1..- bete E1,...')

REPORT DOCUMENTATION PAGE ____RE__

BCT

FORM

I 1IEPORT NUMBER

RADC-TR-75-108

a GOVT ACCESSION NO I RECIPIET'S CATALOG NUMBER

4 TITLE (Od S.te)

S TYPE OF REPORT & PERIOD COVENEO

STUDY AND DEVELGPHENT OF THE INTEL TECHNIQUE FOR

Final 7T3 e-chDneiccal74 Report

S

DiPROVING SFEECH INTELLIGIBILITY

SS PEtRFO7R3MINGDeORcGB.7RR4ErPORT NUMbER

7. AUTHOR(.)

M. R. Weiss E. Aschkenasy

T. W. Parsons

l, PERFORMING ORGANIZATION NAMIE AND ADONE.S

Nicolet Scientifi'? Corporation

145 Livingston Street

Northvale

NJ 07647

I1. CONTROLLING OFFICE NAME AND ADDRESS

Rome Air Development Center (IRAP) Griffiss AFB NY 13441

"* MONITORING AGENCY NAME 4 AODURESS(Iil diflrt

fret Conifiliil

Same

6 CONTRACT OX GRANT MUMBER?*)

F30602-74-C-0058

10. PROGRAM ELEME1T. PROJCT. TASK AREA 4 WORKUNI, NUMBERS

61101F

01737401

12. REPORT OATE

April 1975 111MuSEROFPAGES

$

Office) IS. SECURITY CLASS. (of Thils m rpf)

UNCLASSIFIED

ISe DECL ASSI FICATION, DOWN GRADING

SCH EDULE

S6. DISTRIBUTION STATEMENT (of thle Repomt)

Approved for public release; distribution unlimited.

17. DISTRIBUTION STATEMENT (of IN* .se?rI~t Wfe~ted In Black 20, II different fhw Reopen) Same

I0 SUPPLEMENTARY NOTES

RADC Project Engineer: Robert A. Curtis, Capt, USAF (IRAP)

AC 315 330- 2354 This effort was funded totally by the Laboratory Directors'

II. KEY WORDS (Co.ntinue on *,*tooteSeid* ifoc.neer

#ndiclr ' by block nonib.)

Speech enhancement Speech intelligibility

Fund.

20 ABSTRACT (Cotlln..

,.,*pe .Ide II np.eo e d InlG fy by block .eeoe.)

A process, called INTEL, has been developed for improving the signal-to-.noise

ratio of speech obscured by wideband noise. The orocess teducen the level of

the noise substantially without significantly attenuating or distorting the imbedded speech. At its present stage of development, INTEL's ability to enhance speech intelli?;ibility is uncertain. However, it is effective in reducing listener fatigue, particularly when the S/N Is below 6 dB. The process 1?3ars a superficial similarity to the homomorphic filt.ring techniques

1473 DD oR

EDITION oF 1 NOV 61 IS OsSOLETE

UNCLASSIFIED SECURITY CLASSIFICATION OF THIS PAGE (WhP n Dole Ent.ted)

.-..

UNCLASSIFIED

SIECUNITY CLASSIFICATION OF THIS PAG[f(Wh? DPt itomd)

of Oppenheim et al but differs from them in that it does not use the concept of generalized linearity. It does not appear that this process has been pushed to the limit of its potential. Hence, in the present research, a theoretical analysis of the process has been made, and a number of experiments have been carried out, iu an attemot to improve the performance of the process. The analysis yielded a statistical description of the response of INTEL to white noiFe and to vocalic speech, and a qualitative description of the response to speech and additive noise. The experiments centered about various modifications to the process which seemed likely to render the recovered speech more intelligible and more naturalsounding. The experiments did not lead to distinct improvements in the process. However, the description provided by the analysis pointed the way coward other areas, not covered by the experiments, which look highly promising. In particular, improved methods of removing the noise term in the transformed signal may be successful. It is recommended that these lines be pursued in further research.

id1

UNCLASSIFIED

SECURITY CLASSIFICATION Of THIS PAGE(Wh- Daoe E?i-,d)

TABLE OF CONTENTS

Section 1.0 2.0

3.0

4.0 5 .0

ag~e

Introduction...............................

i

Analysis of the INTEL Process .................... 2

2.1 Description nf Process ................... 2 2.2 Analysis with Noise Input ................ 4 2.3 Comparison with Observed Results ........ 11 2.4 Analysis with Speech Input .............. 13 2.5 Logarithmic Compression ................. 15 2.6 Optimization ............................ 17

Further Developmnent of the INTEL Process ........ 19 3.1 Threshold Clipping ...................... 19 3.2 Pitch Harmonic Emphasis ................. 23 3.3 Processing of Narrow-Band Speech ........ 25

3.4 Investigation of Phase Effects .......... 26

Conclusions and Recoimmendations ............... 30

References .................................. ... 33

Figures...............................*..... . 34

LIST OF FIGURES

i'igure 1. 2. 3.

4. 5.

Block Diagram of INTEL ........................ Simplified Model of INTEL .....................

Probability Density Function of Root-Compressed Noise ....................

Appearance of Pseudo-Cepstrum .................

Probability Density Function of Logarithmically-Compressed Noise .........

Page 34 35

36 37

38

ii

p

LVALUATI ON

This report describes theoretical and experimental studies done to improve the perfornance of INTEL. a process for enhancing the signal-to-noise ratio of speech which has cen corrunted by ividcband noise. The theoretical studies showed some of tile statistical

pronerties of tnc Caussian noise in the INTEL process. .'0hile the intellitil,ilitv of speech is not T1atheraticqlly

definable, several experiments indicate that the rooting tmethod used in part of the INTEI. nrocess tends to separate the speech and noise components.

While the phase is left alone in this process,

experimc?cts indicate taat the correct phase will improve the intelligiLility censiderably. Future work should investigate possible metl.ods of reducing the nhase noise.

The modified ccpstrum gating procezs, as a result of the theoretical ad experimenta.l ork, should also be modified in future research.

The theoretical and experimental york performed

*.

indicated several promising areas for future research.

Capt, USAF Project Engineer

iii

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download