Pulse or digital communications – Receivers – Interference or noise reduction
Reexamination Certificate
1999-07-16
2001-09-18
Chin, Stephen (Department: 2734)
Pulse or digital communications
Receivers
Interference or noise reduction
C370S286000, C381S094100, C704S205000, C704S226000, C704S233000
Reexamination Certificate
active
06292520
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates to a noise canceler to be installed in a telecommunications apparatus designed to encode voice signals for transmission such as a digital cordless telephone set or a digital wired telephone set that is suitably used with a digital portable telephone system or a PCS (personal communication service) system.
A low bit rate voice coding scheme such as the code excited linear prediction (CELP) scheme is popularly used for digital portable telephone sets. With such a coding scheme, voices spoken in an environment with a high background noise level can be clearly heard. The CELP scheme is discussed in detail in M. R. Schroeder and B. S. Atal: “Code-Excided Linear Prediction (CELP): High-Quality Speech at Very Low Bit Rates”, in Proc. ICASSP, 1985, pp. 937-939.
However, spoken voices are remarkably blurred in environments with a high background noise level including buses and commuter trains. Efforts have been made to develop noise cancelers that eliminate noises and encode only voices. Known papers discussing noise cancelers include “Suppression of Acoustic Noise in Speech Using Subtraction” (IEEE rans., vol. ASSP-27, pp. 113-120, April 1979).
The technology discussed in this paper can be summarized as follows. An observed signal is firstly divided into frames with 256 samples and an orthogonal transform operation such as fast Fourier transform is conducted on each of the frames to analyze the frequencies of the signal. Meanwhile, the magnitudes u(i) of the Fourier transform coefficients i of the noise components are observed in advance so that the transform coefficients (i) of the frames of the observed signal may be suppressed by means of the formula below.
S
{circumflex over ( )}(
i
)=max(0
, ∥s
(
i
)∥−
u
(
i
))* sign (
s
(
i
))
Then, the suppressed transform coefficients S{circumflex over ( )}(i) are subjected to an inverse fast Fourier transform (IFFT) to recover the signal, which is subsequently sent to a voice coding section. In this way, the power corresponding to the noise is subtracted from the transform coefficients so that it is theoretically possible to eliminate the noise component from the observed signal and recover the voice.
A constancy is assumed in the transform span for a frequency analysis using orthogonal transform such as FFT. However, the voice is not constant within a frame, nor is the noise if viewed as individual transform coefficient. Both voice and noise can fluctuate with time. Thus, with a known noise canceler that assumes a constancy in the transform span, part of the noise component may remain and/or part of the voice frequencies may be lost or damaged in the noise eliminating operation. These problems then appear as a noise having a specific frequency that can be more annoying than the original sound prior to the noise eliminating operation to baffle the effort for noise suppression.
Generally, an FFT with 256 dimensions is used for a noise canceler. However, an FFT with 256 dimensions involves a large volume of arithmetic operations and hence is not feasibly applicable to small telecommunications apparatus such as portable telephone sets and, therefore, an FFT with dimensions as low as 128, 64 or 32 may have to be used for noise cancelers to be used in small telecommunications apparatus including portable telephone sets. However, an FFT with reduced dimensions is accompanied by a drawback of a long frame length that is longer than the pitch cycles of voice. If a noise eliminating operation is conducted with such a long frame length, assuming a constancy in the transform span as described above, the pitch cycles of voice can be distorted due to the suppressed transform coefficients in a lower frequency band to reproduce a queerly sounding speech if the noise is suppressed effectively.
BRIEF SUMMARY OF THE INVENTION
Therefore, the first object of the present invention is to provide a noise canceler that can effectively eliminate the noise components of a sending speech signal to improve the quality of speech.
The second object of the present invention is to provide a noise canceler that does not distort the voice in a noise suppressing operation if an orthogonal transform with dimensions smaller than the pitch cycles of voice is used so that the noise canceler may operate with a reduced volume of arithmetic operations without degrading the quality of the voice being transmitted.
To achieve the first object, a noise canceler according to the invention divide a sending speech signal by into frames with a constant frame length and carries out an orthogonal transform for frequency analysis on each frame. Then, suppression means of the noise canceler sorts the transform coefficients obtained by the orthogonal transform into a plurality of groups and performs a suppressing operation on the transform coefficients on a group by group basis.
In a first aspect, the suppression means may operate in the following manner. It determines the mean value of the transform coefficients of each of the plurality of groups and compares the mean value of the transform coefficients of each of the plurality of groups with a predetermined threshold value. Then, it uses a minimum value for the transform coefficients of each of the plurality of groups when the mean value of the transform coefficients does not exceed the threshold value and carrying out a suppression processing operation on the transform coefficients according to the absolute values or the square values of the transform coefficients when the mean value of the transform coefficients exceeds the threshold value on the basis of the outcome of the comparison means.
In a second aspect, the suppression means may operate in a manner as described below. Firstly, it determines the mean value of the transform coefficients of each of the plurality of groups and compares the mean value of the transform coefficients of each of the plurality of groups with a predetermined threshold value. Then, it selects a predetermined first value as basic reduction value when the mean value of the transform coefficients exceeds the threshold value and selects a second value that is sufficiently greater than the value as basic reduction value when the mean value of the transform coefficients does not exceed the threshold value. Thereafter, it compares the absolute values of the transform coefficients of each of the plurality of groups and suppresses the transform coefficients of each of the plurality of groups to values obtained by subtracting the basic reduction value from their absolute values and adding a predetermined proportion of each of the absolute values to the outcome of the related subtraction when the absolute values of the transform coefficients of the group exceed the basic reduction value and to values each of which is a predetermined proportion of the absolute value of the related transform coefficient on the basis of the outcome of the second comparison means.
Thus, according to the invention, the transform coefficients obtained by orthogonally transforming the frames of the input voice are treated not as individual coefficients but a set of a plurality of coefficients and divided into groups typically by splitting them according to their frequencies. Then, the coefficients belonging to each of the groups are suppressed strongly if the mean value of the coefficients does not exceed the mean value of the noise levels, whereas they are suppressed only weakly if the mean value of the coefficients exceeds the mean value of the noise levels.
Therefore, with the present invention, the suppressing operation can be carried out under optimum conditions for each of the groups of transform coefficients taking fluctuations in the effect of the transform coefficients into consideration so that the noise components can be effectively eliminated to improve the quality of communication if compared with any known noise cancelers designed to suppress transform coefficients uniformly under same conditions.
The second object of the invention is achieved by providing a no
Chin Stephen
Gray Cary Ware & Freidenrich LLP
Kabushiki Kaisha Toshiba
LandOfFree
Noise Canceler utilizing orthogonal transform does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Noise Canceler utilizing orthogonal transform, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Noise Canceler utilizing orthogonal transform will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2510368