Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1999-11-18
2003-03-25
Dorvil, Richemond (Department: 2654)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S236000, C704S243000, C704S253000
Reexamination Certificate
active
06539350
ABSTRACT:
BACKGROUND OF THE INVENTION
In speech signal processing systems, the current speech level is used, by way of example, for the scaling of signals, for threshold decision, for detection of speech pauses, and/or for automatic adjustment of amplification. Speech level measurement has special significance for successful echo compensation in telecommunications systems, for noise suppression, or in speech recognition in speech coding and speech decoding systems.
The formation of SL (speech level) mean value from sampled values x(k) of a speech signal x(t) within a time interval according to equation G1 is generally known.
SL
=
∑
0
N
&LeftBracketingBar;
x
(
k
)
&RightBracketingBar;
N
(G1)
In the case of speech pauses, the mean value SL assumes the value of the quiescent sound in a period of time determined by the number N of sampled values. At the beginning of the speech activity, a mean value generator requires a period of time determined by the number N to determine the speech level. Determination of a mean value in a time interval of 125 ms requires a data memory of 1000 data words at a sampling rate of 8 kHz. Aside from the considerable computing and memory requirements, in the simple formation of a mean value there is a danger that in the case of a brief averaging period, errors will occur in determining the speech level as a result of interference factors. In the case of long averaging periods, first the information concerning the value of the speech level is available very late, and secondly measuring errors with respect to the speech level occur in the event of changes in speech level.
Also known is the use of recursive filters for the formation of a mean value; compare Hentschke:
Grundzüge der Digitaltechnik
(Fundamentals of Digital Technology), Stuttgart: Teubner 1988, pages 52-54. The computing and memory requirements for these digital filters are relatively small; however, all signal values are determined so that distinguishing between speech and interference noise is not possible.
From the field of speech processing, the method of linear prediction (linear predictive coding, LPC) is known with which distinguishing features of speech and interference noise can fundamentally also be determined. LPC analysis is very precise and can be performed very quickly and is a powerful method with which, among other things, the base frequency, spectrum, and formats of a speech signal can be determined; compare Eppinger, Herter:
Sprachverarbeitung
(Speech Processing), Munich, Vienna: Hanser 1983, pages 73-77. Such a costly method, however, is not suitable for mass products such as telecommunications terminal devices for commercial reasons.
SUMMARY OF THE INVENTION
The invention solves the object of suggesting a cost-effective, practicable method for speech level measurement and a circuit arrangement for implementing the method having the following properties:
From a time signal the current speech level is to be determined as quickly and precisely as possible,
The adaptation period of the speech level measurement circuit should be short in order to avoid audible errors such as fluctuations in loudness,
The measured speech level should be independent of level fluctuations of the speech caused, for example, by nasal sounds and open vowels,
The measured speech level should be independent of short-time disturbance influences such as, for example, whispering, coughing, clapping, slamming of doors, although these particular interferences have a high energy content,
In speech pauses, the measured value of the speech level should be maintained in order to suppress the breathing of loudness known from automatic gain control, AGC.
This object is achieved through the method described in the first patent claim and through the circuit arrangement described in the seventh patent claim. The essence of the invention consists of a measured speech level value being admitted for further processing in a speech signal processing system only if characteristic features of speech are recognized and interference signals and speech pauses being filtered out for the measurement.
REFERENCES:
patent: 4032710 (1977-06-01), Martin et al.
patent: 4625083 (1986-11-01), Poikela
patent: 4625327 (1986-11-01), Sluijter et al.
patent: 4637046 (1987-01-01), Sluijter et al.
patent: 4696039 (1987-09-01), Doddington
patent: 5305422 (1994-04-01), Junqua
patent: 32 36 834 (1983-10-01), None
patent: 32 30 391 (1984-02-01), None
patent: 689 03 872 (1993-06-01), None
patent: 0 565 224 (1994-10-01), None
patent: 691 05 154 (1995-03-01), None
patent: 07 326 981 (1995-12-01), None
Gansler et al (“Non-Intrusive Measurements of the Telephone Channel”, IEEE Transactions on Communications, Jan. 1999).*
Bertocco et al (“In-Service Non-Intrusive Measurement of Noise and Active Speech Level in Telephone-Type Networks”, IEEE Transactions on Instrumentation and Measurement, Aug. 1998).*
Bentelli et al (“A Multi-channel Speech/Silence Detector based on Time Delay Estimation and Fuzzy Classification”, IEEE International Conference on Acoustics, Speech, and Signal Processing, Mar. 1999).*
McKinley et al (“Model Based Speech Pause Detection”, IEEE International Conference on Acoustics, Speech, and Signal Processing, Apr. 1997).*
Ramsden (“In-Service, Non-Intrusive Measurement on Speech Signals”, Global Telecommunications Conference on Personal Communications Services, May 1991).*
Hentschke: “Grundzuge der Digitaltachnik (Fundamentals of Digital Technology)”, Stuttgart: Teubner 1988, pp. 52-55.
Eppinger, Herter: “Sprachverarbeitung (Speech Processing)”, Munich, Vienna: Hanser 1983, pp. 73-77.
Bauer, B. B. et al.: “The Measurement of Loudness Level” Journal of the Acoustical Society of America, US, American Institute of Physics. New York, BD. 50, Nr. 2, Part 01, Aug. 1971, pp. 405-414 XP000795762 ISSN: 0001-4966.
Alcatel
Dorvil Richemond
Nolan Daniel A.
LandOfFree
Method and circuit arrangement for speech level measurement... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and circuit arrangement for speech level measurement..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and circuit arrangement for speech level measurement... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3069214