JPH11338499A - Noise canceller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To eliminate a defect of a speaking interruption caused by elimination of one part of a sound signal in a conventional noise canceller when used under an environment with a high level of a background noise. SOLUTION: A noise is suppressed by a noise canceller A1 using Kalman filter algorithm adaptable to a low level of a background noise when it is in the low level, and the noise is suppressed by a noise canceller B2 using a speech processing system (SPAC) utilizing an autocorrelation function adaptable to a high level of the background noise when it is in the high level.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise canceller used in digital voice communication or digital audio (sound) recording, and more particularly, to a noise canceller used in an environment having a large background noise.

[0002]

2. Description of the Related Art As an example of the prior art, a PDC system (Pers
A noise canceller used in a mobile phone of the onal Digital Cellular Telecommunication System) will be described. FIG. 4 is a configuration diagram of a speech encoding apparatus to which the present invention is applied. In FIG. 4, an input signal a is a digital signal. When background noise is superimposed on voice, the background noise is removed by a noise canceller 11, only a voice signal is extracted, and the voice coder 12 converts the voice signal into coded voice data. Is done. Here, the input signal a is, for example,
This is a digital signal sequence sampled at 8 kHz and quantized at 16 bits.

[0003] The speech encoder 12 has a signal speed of 1
If the full rate is 1.2 kbps, VSELP (Vector S
um Excited Linear Prediction) encoder is used, and if the signal rate is 5.6 kbps at half rate, PS
I-CELP (Pitch Synchronous Innovation CEL)
An encoder based on P, CELP (Code Excited Linear Prediction) is used. For reference, the details of these audio coding processes are described in the standard STD-27C of the Radio System Development Center (RCR).

FIG. 5 is a diagram showing an example of the configuration of a conventional noise canceller 11. In the figure, 13 is a state determiner, 14 is an input parameter calculator, and 15 is a Kalman filter.
Table 1 is a table in which the degree of superimposition of the background noise is classified. As an input signal, a signal a in which background noise is superimposed on voice is input to the state determiner 13. The state determiner 13 previously determines the six states (11, 10, 0,
20, 21, 22), and determines which state the input signal a is in and outputs it. The noise removal operation in Table 1 shows the operation of the Kalman filter 15.

[0005]

[Table 1]

[0006] The input parameter calculator 14 determines the state of the output of the state determiner 13 according to the state of the Kalman filter 1.
5 calculates and outputs a parameter that gives a cancel strength so that the noise removal operation (Table 1) according to the state (11, 10, 0, 20, 21, 22) is performed. The Kalman filter 15 removes the background noise component from the signal a in which the background noise is superimposed on the voice by using the parameter indicating the cancellation strength output from the input parameter calculator 14, and outputs the resulting voice signal.

[0007]

However, the above-mentioned conventional circuit works well when the background noise is small, but when the portable terminal is used in a noisy environment such as a construction site.
That is, when used in an environment where the power of the background noise is larger than the power of the audio signal, for example, in an environment where the signal-to-noise ratio (S / N) <0 dB, the noise canceller using the Kalman filter algorithm of the prior art shown in FIG. In addition, since not only the background noise but also a part of the audio signal is removed, there is a problem that the audio signal output from the noise canceller has a noticeable discontinuity and is difficult to hear.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problem, and provides a noise canceller capable of removing a noise component without interrupting a voice signal even in an environment having a large background noise. Is to do.

[0009]

SUMMARY OF THE INVENTION A noise canceller of the present invention removes background noise superimposed on a sampled and quantized input signal for input to a speech coder or digital sound recorder. For this purpose, a plurality of types of noise cancellers having different cancellation strengths are switched and used according to the level of background noise superimposed on the input signal.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The noise canceller of the present invention uses a noise canceller having a strong cancellation strength by a speech processing system (SPAC) using an autocorrelation function when the background noise is large, and a weak cancellation strength when the background noise is small. By switching and using the Kalman filter algorithm type canceller, it is possible to obtain a sound signal or sound recording with a reduced noise level regardless of the magnitude of the background noise, and especially to prevent the sound signal from being interrupted when the background noise is large. it can.

Here, a speech processing system using an autocorrelation function (SPAC: Speach Processing system by use of
For details of the Auto Correlation function, see the paper by Seiji Suzuki et al. (“Speech processing method using autocorrelation function (SPA)
Function and basic characteristics of C) "IEICE, J62-A, No.3, pp.1
75-182, 1979, March).

A correlation technique using an autocorrelation function is a technique used for signal analysis and feature extraction, statistical analysis of noise, separation of two or more signals, detection and measurement of a periodic signal, and the like. In the field of voice processing, (1) voice analysis,
(2) High-efficiency transmission of voice, and (3) Reduction of noise level superimposed on voice signal.

The autocorrelation function φ of the periodic wave has the same frequency component as the original signal, and the period can be easily detected. on the other hand,
The amplitude component of the random noise φ is concentrated near the origin. The speech processing system SPAC uses the property of φ to regard the short-time autocorrelation function of speech as a waveform.
This is a method of playing back audio signals by connecting periodic waveforms.
The noise level can be reduced and the SN ratio can be improved by using the difference in the correlation function between the noise and the signal. When the noise reduction processing by the SPAC is performed on the quantized signal, the noise of the pause is largely suppressed, and it becomes very easy to hear.

The noise cancellation system using the Kalman filter algorithm has been described in the prior art, and is described in detail in RCR STD-27C.

Table 2 is a comparison table of the properties and characteristics of these noise canceling systems.

[0016]

[Table 2]

According to Table 2, the switching level is about 5 dB S / N, the Kalman filter algorithm is used in an environment where S / N> 5 dB, and the SP is used in an environment where S / N <5 dB.
By switching and using each noise canceller like using AC, it is possible to improve the communication quality.

As a method of switching the noise canceling method, two methods, a manual method and an automatic method, are effective. A method of manually switching a switch on the operation panel while making a call.
Alternatively, an automatic switching method using a noise canceller selector that calculates a background noise level (or S / N) and automatically selects one of two types of noise cancellers respectively suitable for the background noise level (or S / N). There is.

[0019]

FIG. 1 is a block diagram showing a first embodiment of the present invention, and shows a configuration example in which a switch on an operation panel is manually switched as a switching method of a noise canceling method. In the figure, 1 is a noise canceller A using a Kalman filter algorithm, and 2 is a noise canceller B using SPAC. Reference numerals 3 and 4 denote switching units, and reference numeral 5 denotes an audio encoder.

The signal a in which the background noise is superimposed on the voice is switched by the switch 3 so as to be input to either the noise canceller A1 or the noise canceller B2.
The noise canceller A1 uses a Kalman filter algorithm and is effective when there is little background noise. The noise canceller B2 is a system using SPAC, and is effective when background noise is large.

The switch 4 works in conjunction with the switch 3 to switch between the output of the noise canceller A1 and the output of the noise canceller B2, and inputs the selected signal to the speech encoder 5.
The audio encoder 5 performs an encoding process on the output from the switch 4,
The resulting encoded audio data is output.

The switching control signal f is a noise canceller A
1 and a signal for switching the switch 3 and the switch 4 in an interlocking manner so as to select one of the noise canceller B2, and is configured to be generated by a switch operation on an operation panel of the audio encoding apparatus.

FIG. 2 is a block diagram showing a second embodiment of the present invention. The background noise level (or the S / N of the input signal) is calculated, and the background noise level (or the S / N of the input signal) is calculated. A configuration for automatically selecting a suitable noise canceller will be described. The difference from the configuration of the first embodiment of the present invention shown in FIG. 1 is that only the method of generating the switching control signal (f in FIG. 1 and g in FIG. 2) is different. Hereinafter, that portion will be described. The other operations are the same as those of the first embodiment shown in FIG.

The switching control signal g is generated by the selector 6. The processing in the selector 6 will be described below. The selector 6 generates a switching control signal g for each frame. Where 1
The length of the frame is set equal to the frame length of the encoding process of the audio encoder. For example, it is set to 40 msec.

FIG. 3 is a flowchart of an automatic switching process of the selector 6 according to the second embodiment of the present invention. This flowchart shows the processing performed for each frame. However, as an initial setting, the counter in the selector 6 is set to 0.
Is set. 3 in FIG. 3 indicate step numbers. The “step” selector 6 calculates a frame power P of the input signal a for each frame. [Step] The moving average value Pm of the frame power over the past m frames from the current frame is calculated and updated (for example, m = 10).

[Step] The power value P of the current frame with respect to the moving average value Pm of the power of the past (m) frames
, That is, P / Pm is greater than or equal to threshold 1 (th1),
If it is equal to or less than the threshold value 2 (th2), the process proceeds to step and adds 1 to the counter (for example, th1 = 0.75, t
h2 = 1.25). If it does not fall within this range, the process proceeds to step and the counter is set to 0. This process is based on the following idea. P / Pm is the current frame power P
Is the ratio of the power of the past m frames to the moving average Pm. In the background noise section (no voice signal exists), the fluctuation of the frame power is small (P / Pm becomes a value close to 1). Therefore, if several frames in which P / Pm is equal to or more than the threshold value 1 (1th) and equal to or less than the threshold value 2 (th2) continue several times, it can be estimated that the current frame is a background noise section.
This continuous number is checked using a counter.

If the counter value indicating the "step" frame continuation number is equal to or larger than the threshold value 3 (th3), it is determined that the current frame is a background noise section (for example, th3 = 5). When it is determined in the “step” step that the current time is in the background noise section, the process proceeds to the step and the frame power
And the threshold value 4 (th4). If the difference is equal to or less than th4, the process proceeds to step S1 to select the noise canceller A1 (a method using a Kalman filter algorithm).
C), the state of the switching control signal g is switched and output. Here, th4 is, for example,
Set to 1/4 of the frame power at full scale.

The first and second embodiments of the present invention have been described above, but both are examples of voice coded communication terminals such as digital portable telephones. However, if the condition of the input signal a, that is, if the input signal is a digital signal sequence sampled at 8 to 10 kHz and quantized at 11 to 16 bits, a digital recorder is provided instead of the voice encoder 5. Then, the present invention can be applied to digital audio equipment, and the same effect can be obtained.

Further, as the above examples, there are two methods, a method using the Camarn filter algorithm and a method using SPAC.
An example of the configuration for switching between different types of noise cancellers has been described. For example, the spectral subtraction method (S
It goes without saying that other types of noise cancellers such as the S method) may be combined and arranged.

[0030]

As described above, by implementing the present invention, when a voice communication device or a sound recording device is used in a noisy environment such as a construction site or an aircraft, for example, the power of background noise is reduced. When used in an environment where the power of the audio signal is larger than the power of the audio signal (signal-to-noise ratio (S / N) <0 dB), noise can be effectively removed, and the quality of voice and sound recording can be improved.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a first embodiment of the present invention.

FIG. 2 is a configuration diagram showing a second embodiment of the present invention.

FIG. 3 is a flowchart of an automatic switching process according to the present invention.

FIG. 4 is a block diagram of a speech coding apparatus to which the present invention is applied.

FIG. 5 is a diagram illustrating a configuration example of a conventional noise canceller.

[Explanation of symbols]

Reference Signs List 1 noise canceller A using Kalman filter algorithm 2 noise canceller B using SPAC 3,4 switch 5 audio encoder 6 selector 11 noise canceller 12 audio encoder 13 state determiner 14 input parameter calculator 15 Kalman filter

Claims (4)

[Claims] 1. A method for removing background noise superimposed on a sampled and quantized input signal for input to a speech coder or a digital sound recorder, wherein said superimposed signal is superimposed on said input signal. A noise canceller configured to switch and use a plurality of types of noise cancellers having different cancellation strengths according to the level of the background noise. 2. In order to remove background noise superimposed on a sampled and quantized input signal for input to a speech coder or a digital sound recorder, the signal is superimposed on the input signal. A first noise canceller based on a speech processing method using an autocorrelation function having a strong canceling strength against background noise, and a second noise canceller based on a method using a Kalman filter algorithm having a weak canceling strength on background noise superimposed on the input signal When the level of the background noise superimposed on the input signal is greater than a preset threshold, the first noise canceller suppresses the background noise, and the level of the background noise superimposed on the input signal is When the value is smaller than the threshold, background noise is suppressed by the second noise canceller and output. A noise canceller comprising a switching means for switching as described above. 3. The noise canceller according to claim 1, wherein said switching means is configured to be manually operated. 4. The switching means includes a selector for calculating a background noise level superimposed on the input signal for each frame, and selecting a noise canceller according to the calculated background noise level. 3. The apparatus according to claim 1, wherein a suppressed output is obtained.
The described noise canceller.