JPH0583039B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an acoustic noise removal device, and more particularly to an acoustic noise removal device that has two acoustic signal reception points and removes noise components.

[Conventional technology]

Using two acoustic detection elements such as microphones,
Acoustic noise removal devices that analyze the characteristics of the environmental noise component captured by one of the microphones and then use this analysis result to remove the environmental noise component from the input of the other microphone are being used frequently these days. It's coming.

The above-mentioned noise environment is mainly intended for cases where extremely high level noise is received, such as inside various ships or aircraft, and therefore, the noise has a harmonic structure generated by rotating machines, etc. It is not uncommon for the level of communication to reach such a level that it is impossible to transmit and receive voice communications.

Conventionally, in order to remove such noise, waveform domain processing in which the noise is removed using a filter has been frequently used.

[Problem that the invention seeks to solve]

However, the propagation path of acoustic noise in the environment described above is generally complex, and the path changes from time to time, so-called time-varying.
Variant), and there is a problem in that it is difficult to perform effective noise removal in such a state.

This problem arises because the acoustic noise processing algorithm does not function well when the time-varying speed of the acoustic noise propagation path is faster than the time required to estimate the acoustic noise propagation path.

Another object of the present invention is to eliminate the above-mentioned drawbacks and to provide an acoustic noise removal device that functions well against acoustic noise having a harmonic structure even when the time-varying speed of the propagation path is fast.

[Means for solving problems]

The device of the present invention is an acoustic noise removal device having two acoustic signal receiving points, and detects the fundamental frequency (or period) of a component having a harmonic structure in the acoustic signal received at one acoustic signal receiving point. The fundamental frequency detecting means calculates the frequency spectrum of the acoustic signal received at another acoustic signal receiving point, and the fundamental frequency detected by the fundamental frequency detecting means with respect to this frequency spectrum and a frequency having a harmonic relationship therewith. and comb filtering implementation means for attenuating the components by comb filtering in the frequency domain.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.

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

The configuration of the embodiment shown in FIG.
2, a fundamental frequency detection section 3, a frequency spectrum calculation circuit 4, a frequency domain comb filtering circuit 5, and the like.

Further, FIG. 2 is a block diagram showing in detail the fundamental frequency detecting section 3 in the embodiment of FIG. ,
Autocorrelation coefficient calculator 35 and peak value searcher 3
6.

The microphone 1 is placed at an acoustic signal receiving point where environmental noise is predominant, and inputs the environmental noise. After this environmental noise is converted into an electrical signal, it is supplied to the fundamental frequency detection section 3 to detect the fundamental frequency component of the environmental noise having a harmonic structure, and the data related to this is sent to the frequency domain comb filtering circuit 5. supply

On the other hand, necessary voices are input to the microphone 2 along with environmental noise, which are converted into electrical signals and then supplied to the frequency spectrum calculation circuit 4.

The frequency spectrum calculation circuit 4 performs frequency spectrum calculation for converting a time domain input signal into a frequency domain signal.
This is done using a Fast Fouriers Transform (Fouriers Transform) circuit. The noise and speech thus converted into the frequency domain are supplied to the frequency domain comb filtering circuit 5.

Now, the fundamental frequency detection section 3 detects the fundamental frequency component of a noise series of noise captured by the microphone 1 in an operational environment that has a harmonic structure, and in this embodiment, this is shown in FIG. Processing is performed according to the content shown. Hereinafter, the detection contents of the fundamental frequency will be explained based on FIG. 2.

An input signal input from the microphone 1 is supplied to a window processor 31 and subjected to window processing by multiplication with a predetermined window function. In this case, the input signal is sampled at a predetermined sampling frequency, and then the sampled samples are quantized using a predetermined number of bits. The sampling frequency and the number of quantization bits can be set arbitrarily, taking into account the maximum frequency to be quantized, the quantization accuracy, etc., respectively.

The quantized signal is then successively extracted at a predetermined time length through convolution multiplication with a rectangular function or a window function such as a Hamming function, and is supplied to the Fourier transformer 32.

The Fourier transformer 32 converts the input quantized signal into digital quantities in the frequency domain via a DFT (Discrete Fouiers Transform) circuit, and supplies these to the power spectrum calculator 33.

The power spectrum calculator 33 calculates a power spectrum of a scalar amount through square addition of the real part and imaginary part of the frequency spectrum components of the input signal, and temporarily stores them in an internal memory.

The power spectrum of each window function extraction unit stored in the internal memory of the power spectrum calculator 33 is then read out one after another for a predetermined length of time to the data nonlinear compressor 34.

The data nonlinear compressor 34 performs nonlinear compression of the input scalar power spectrum data. This nonlinear compression is intended to greatly facilitate and improve the accuracy of the evaluation of the input power spectrum.

There are various types of nonlinear compression of the input, such as nonlinear compression such as logarithmic compression or mu-curve compression, or center clipping, which compresses the input by clipping it with a level below a preset slice level. In the example, logarithmic compression is performed and the data is supplied to the autocorrelation coefficient calculator 35.

The autocorrelation coefficient calculator 35 calculates an autocorrelation coefficient sequence in a necessary delay time range for the nonlinearly compressed input data and sends it to the peak value detector 36.

The peak value detector 36 searches for the peak value of the input autocorrelation coefficient, and calculates the delay of the autocorrelation coefficient corresponding to this peak value, that is, the delay on the frequency axis, and more specifically, the delay of the autocorrelation coefficient that corresponds to this peak value. In other words, the fundamental frequency of environmental noise having a harmonic structure can be easily determined.

Let us return to Figure 1 and continue the explanation.

The frequency domain comb filtering 5 nulls the fundamental frequency and its harmonic components supplied from the fundamental frequency detection section 3, that is, frequency components having a harmonic structure, with respect to the frequency spectrum input from the frequency spectrum detection circuit 4. Apply comb filtering (null) to the pole in the frequency domain.

FIG. 3 is a frequency domain comb filtering attenuation characteristic diagram showing the characteristics of the frequency domain comb filtering in the embodiment of FIG.

The frequency domain comb filtering circuit 5 includes a comb filter having a comb filtering characteristic indicated by P, and filters the frequency spectrum input from the frequency spectrum calculation circuit 4. In addition to the fundamental frequency f ₀ , the above-mentioned type filter has 2f ₀ ,
3f ₀ . . . Generally, a sharp attenuation pole, that is, a null pole, is set at the frequency position nf ₀ . nf ₀ can be set arbitrarily considering the upper limit of the noise frequency to be removed, and the attenuation bandwidth with 3 dB (devesyl) reduction
b ₀ can also be made as sharp as desired. By passing through a filter with such attenuation characteristics, noise with a harmonic structure can be attenuated accurately and sufficiently, and furthermore, the frequency components of the harmonic structure will always remain correct even under the time change of the acoustic noise propagation path. Captured noise can be removed.

The fundamental frequency detection unit 3 in the embodiment shown in FIG. 1 described above has the configuration shown in FIG. The basic method is to search for the peak value of a sequence and remove the noise component of a harmonic structure with repetitive periodicity, but various other methods have been considered based on similar ideas to detect the fundamental frequency by focusing on periodicity. It will be done.

For example, a detection method using cepstrum analysis is one such method.

In other words, the cepstrum obtained by calculating the power spectrum from DFT of the input signal, taking its logarithm, and DFT again will have a peak at a quefrency equal to the period if the original signal has a harmonic structure and is periodic. This can be used to accurately determine the fundamental frequency.

〔Effect of the invention〕

As explained above, according to the present invention, under environmental noise propagating through a time-varying propagation path, the fundamental frequency of a noise component having a harmonic structure is detected, and the fundamental frequency and its harmonic relationship are detected. By inputting an acoustic signal through a comb-shaped filter that has frequency components as attenuation poles and removing periodic environmental noise, an acoustic noise removal device with significantly improved noise removal effects can be realized There is an effect.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention;
FIG. 2 is a block diagram showing in detail the fundamental frequency detecting section 3 in the embodiment shown in FIG. 1, and FIG. 3 is a frequency domain comb filtering attenuation characteristic diagram in the embodiment shown in FIG. 1, 2...Microphone, 3...Fundamental frequency detection section, 4...Frequency spectrum calculation circuit, 5...
Frequency domain comb filtering circuit, 31...
Window processor, 32...Fourier transform unit, 33...Power spectrum calculator, 34...Data nonlinear compressor, 35...Autocorrelation coefficient calculator, 36...Peak value detector.

Claims (1)

[Claims] 1. An acoustic noise removal device having two acoustic signal receiving points, and a fundamental frequency detection means for detecting the fundamental frequency (or period) of a component having a harmonic structure among the acoustic signals received at one acoustic signal receiving point. Then, the frequency spectrum of the acoustic signal received at another acoustic signal receiving point is calculated, and the fundamental frequency detected by the fundamental frequency detecting means and the frequency component having a harmonic relationship with this frequency spectrum are calculated. 1. An acoustic noise removal device comprising: comb filtering implementation means for providing attenuation by comb filtering in a target frequency region.