JP7630872B2 - Noise Update Circuit - Google Patents

Description

This invention relates to a noise update circuit, for example, a noise update circuit that can be used in a noise reduction circuit incorporated in a radio device that transmits and receives high-frequency signals.

Spectral subtraction is known as a method for suppressing noise components contained in an audio signal (see, for example, Patent Document 1 and Non-Patent Document 1).

Japanese Patent Application Publication No. 4-238399

P. Scalart and J. Vieira Filho “Speech Enhancement Based on a Priori Signal to Noise Estimation”, IEEE International Conference on. Acoustics, Speech, Signal Processing, Atlanta, GA, USA, vol. 2, pp. 629-632, 1996

However, in order to properly apply the spectral subtraction method, it is important to accurately estimate the noise components and subtract them from the input speech signal.

Therefore, the objective of this invention is to provide a noise update circuit that can accurately estimate noise components.

In order to solve the above problem, the present invention provides a first updating unit that, when a frame to be processed is a frame containing only noise components, calculates an updated noise spectrum Ni(f) for each frequency f in accordance with the following Equation 1 which is an IIR filter or the following Equation 2 which is an FIR filter, using a signal corresponding to an amplitude spectrum of the frame to be processed as an input signal spectrum Y(f) (where Ni-1(f): noise spectrum one frame before the update, Yi-j(f): input signal spectrum j frames before the update, Kn: constant determining weighting of Ni-1(f) with respect to Yi(f) when the frame to be processed is a frame containing only noise components, i: ordinal number representing the order in time series, j: integer equal to or greater than 0 representing the degree of distance from ordinal number i in the time series);
a second update unit that calculates an updated noise spectrum N(f) for each frequency f according to the following Equation 3 which is an IIR filter or the following Equation 4 which is an FIR filter when the frame to be processed is a frame including a voice component (where Ks is a constant that determines the weighting of N(f) with respect to Y(f) when the frame to be processed is a frame including a voice component, Kn<Ks);
an average calculation unit that calculates an average spectrum level M for the input signal spectrum Y(f) according to the following formula 5 (where f: minimum frequency in the amplitude spectrum, f: maximum frequency in the amplitude spectrum, F: number of frequencies in the range from the minimum frequency f to the maximum frequency f);
a third update unit that determines an updated noise spectrum Ni(f) according to the following Equation 6 for a frequency f where Yi(f)≧Te×M with respect to the input signal spectrum Yi(f) and the average spectrum level M (where T is a coefficient, and Ni-1(f) is the noise spectrum one frame before the update);
The noise updating circuit further comprises :

The invention described in claim 2 is characterized in that, in the noise updating circuit described in claim 1 , the constant Ks is set to any value within a range corresponding to a time constant of an IIR filter or an averaging interval of an FIR filter, which is 1 to 10 seconds.

According to a third aspect of the present invention, in the noise updating circuit according to the first aspect, the coefficient Te is set to any value within a range of 1 to 100.

According to the invention described in claim 1, even if frames containing voice components are consecutively processed as frames to be processed, it is possible to follow the noise fluctuations and accurately estimate the noise components. Specifically, in a conventional circuit that realizes the spectrum subtraction method, the noise spectrum is updated when the frame to be processed is a frame containing only noise components, but the noise spectrum is not updated when the frame contains voice components. Therefore, if frames containing voice components are consecutively processed as frames to be processed, the noise spectrum is not updated, so it is not possible to follow the noise fluctuations accurately, and as a result, it is not possible to accurately estimate the noise components. In contrast, according to the invention described in claim 1 , the noise spectrum is updated even when the frame to be processed is a frame containing voice components, so it is possible to follow the noise fluctuations and accurately estimate the noise components even if frames containing voice components are consecutively processed as frames to be processed. Also, according to the invention described in claim 1, it is possible not to suppress the tone signal. Specifically, in conventional circuits that realize the spectral subtraction method, components that are constantly present in the frequency spectrum are judged to be noise and suppressed, so that tone signals, which are also stationary, are judged to be noise and are subject to suppression, resulting in the problem that tone signals necessary for the user (for example, Morse code) are also suppressed. In contrast, in the invention described in claim 1, frequency components whose amplitude spectrum is significantly larger than the average spectrum level are excluded from updating of the noise spectrum, making it possible to avoid suppressing tone signals.

According to the invention recited in claim 2 , the constant K determining the weighting of the noise spectrum N(f) of the previous frame to be updated for the input signal spectrum Y(f) when the frame to be processed is a frame containing a speech component is set to an appropriate value. Therefore, it is possible to more reliably follow noise fluctuations when frames containing speech components are successively processed as frames to be processed, and it is possible to more reliably and accurately estimate the noise components.

According to the invention recited in claim 3 , the coefficient Te is set to an appropriate value, so that it is possible to more reliably exclude frequency components whose amplitude spectrum is significantly larger than the average spectrum level from the updating of the noise spectrum, and it is possible to more reliably prevent tone signals from being suppressed.

1 is a functional block diagram showing a schematic configuration of a noise reduction circuit including a noise updating circuit according to an embodiment of the present invention; 1 is a functional block diagram showing a schematic configuration of a noise updating circuit according to a first embodiment; FIG. 11 is a functional block diagram showing a schematic configuration of a noise updating circuit according to a second embodiment.

The present invention will be described below based on the illustrated embodiment.

(Embodiment 1)
Fig. 1 is a functional block diagram showing a schematic configuration of a noise reduction circuit 1 including a noise updating circuit 11 according to an embodiment of the present invention. Fig. 2 is a functional block diagram showing a schematic configuration of the noise updating circuit 11 according to the first embodiment.

The noise reduction circuit 1 is a circuit that is incorporated, for example, in a radio that transmits and receives high-frequency signals, and realizes the Spectral Subtraction method, which is a method for suppressing noise components contained in audio signals. It mainly has a pre-emphasis circuit 2, a window processing unit 3, a time-frequency conversion unit 4, a conversion result output unit 5, a subtraction unit 6, a synthesis unit 7, a frequency-time conversion unit 8, a de-emphasis circuit 9, a voice section detection unit 10, and a noise update circuit 11.

The pre-emphasis (PE) circuit 2 performs high-frequency emphasis processing to pre-amplify the relative intensity of high-frequency components on the audio signal demodulated from the high-frequency signal received from the antenna, and outputs the signal after high-frequency emphasis processing.

The window processing unit 3 receives the high-frequency emphasis processed signal output from the pre-emphasis circuit 2, extracts frames of a predetermined length from the input signal (e.g., extracts a 25 ms time waveform every 12.5 ms), and performs window processing on each frame by multiplying it by a window function such as a Hanning window. Each time the window processing unit 3 performs window processing on a frame, it outputs the windowed frame.

The time-frequency conversion unit 4 receives the windowed frame output from the window processing unit 3, and each time the time-frequency conversion unit 4 receives the frame, it converts the frame from a time domain signal to a frequency domain signal, calculates a frequency spectrum including amplitude and phase components for each of a plurality of frequencies, and outputs a signal with a real and imaginary frequency spectrum. The time-frequency conversion unit 4 performs a time-frequency conversion, for example, by a discrete Fourier transform or a fast Fourier transform, to calculate the frequency spectrum.

The conversion result output unit 5 receives the frequency spectrum signal for each frame (e.g., at intervals of about 12.5 ms) output from the time-frequency conversion unit 4, and outputs, for each frame, a signal corresponding to an amplitude spectrum including the amplitude components of each frequency in the input frequency spectrum to the subtraction unit 6, and outputs a signal corresponding to a phase spectrum including the phase components of each frequency in the input frequency spectrum to the synthesis unit 7.

The subtraction unit 6 receives an input of a signal corresponding to the amplitude spectrum for each frame output from the conversion result output unit 5, and also receives an input of a signal corresponding to the updated noise spectrum for each frame output from the noise update circuit 11, and subtracts the signal corresponding to the input updated noise spectrum for each frequency (in other words, for each spectrum) from the signal corresponding to the input amplitude spectrum for each frame. This suppresses the noise components contained in the audio signal. The subtraction unit 6 outputs a signal corresponding to the amplitude spectrum after subtraction for each frame output from the conversion result output unit 5.

The synthesis unit 7 receives an input of a signal corresponding to the phase spectrum for each frame output from the conversion result output unit 5, and receives an input of a signal corresponding to the amplitude spectrum after subtraction processing for each frame output from the subtraction unit 6, synthesizes the input signal corresponding to the phase spectrum and the signal corresponding to the amplitude spectrum for each frame to generate a frequency spectrum, and outputs a signal with a frequency spectrum of real and imaginary numbers.

The frequency-time transform unit 8 receives the frequency spectrum signal for each frame output from the synthesis unit 7, and performs a conversion process from a frequency domain signal to a time domain signal for each frame on the input frequency spectrum signal, i.e., an inverse conversion process of the conversion process in the time-frequency transform unit 4, to output an audio signal. The frequency-time transform unit 8 performs a frequency-time conversion, for example, by an inverse discrete Fourier transform or an inverse fast Fourier transform, to generate an audio signal.

The de-emphasis (DE) circuit 9 receives the audio signal output from the frequency-time converter 8, performs high-frequency attenuation processing to attenuate the relative intensity of high-frequency components in the input audio signal, i.e., attenuation processing using the inverse filter of the pre-emphasis circuit 2, and outputs the audio signal after high-frequency attenuation processing.

The speech section detection unit 10 receives the signal corresponding to the amplitude spectrum for each frame output from the conversion result output unit 5 and branches off, and determines for each frame whether the signal corresponding to the input amplitude spectrum is a frame containing only noise components or a frame containing speech components.

The method used by the voice activity detection unit 10 to determine whether a frame being processed contains only noise components or also contains voice components is not limited to a specific procedure or method, but rather an appropriate procedure or method may be selected from conventional or new procedures or methods.

As a method for determining whether a frame to be processed contains only noise components or includes speech components in the voice section detection unit 10, for example, a method that focuses on the non-constancy of speech and determines that the frame to be processed contains only noise components if the average or variance value of the amplitude magnitude for each frequency of the amplitude spectrum is less than a predetermined threshold value multiple times (e.g., about 3 to 5 times) consecutively in the most recent frame, and determines that the frame to be processed contains speech components in other cases, or a method that determines that the frame to be processed contains only noise components if the average or variance value of the amplitude magnitude for each frequency of the amplitude spectrum is less than a predetermined threshold value, and determines that the frame to be processed contains speech components if the average or variance value is equal to or greater than the threshold value.

The voice section detection unit 10 outputs a noise frame signal when it determines that the frame to be processed contains only noise components, and outputs a voice frame signal when it determines that the frame to be processed contains voice components. The voice section detection unit 10 outputs a noise frame signal or a voice frame signal as the voice section detection result for each frame.

The noise update circuit 11 according to the embodiment has a first update unit that calculates an updated noise spectrum Ni(f) for each frequency f by processing including averaging using a signal corresponding to the amplitude spectrum of the frame to be processed as the input signal spectrum Yi(f) when the frame to be processed is a frame containing only noise components (where i is an ordinal number representing the order in the time series), and a second update unit that calculates an updated noise spectrum Ni(f) for each frequency f by processing including averaging when the frame to be processed is a frame containing voice components, such that the averaging time of the input signal spectrum Yi(f) when calculating the updated noise spectrum Ni(f) when the frame contains voice components is longer than the averaging time of the input signal spectrum Yi(f) when calculating the updated noise spectrum Ni(f) when the frame contains only noise components.

The noise update circuit 11 updates the noise spectrum representing the noise components for each frequency calculated in the past to the latest noise spectrum by adding the amplitude spectrum of the current frame (in other words, the frame to be processed, the latest frame), and has a first update unit 111 and a second update unit 112.

The noise update circuit 11 receives an input of a signal corresponding to the amplitude spectrum for each frame output from the conversion result output unit 5 and branched, and also receives an input of the voice activity detection result for each frame output from the voice activity detection unit 10, and updates the noise spectrum Ni-1(f) of the previous frame for each frequency f using the input signal corresponding to the amplitude spectrum, and calculates the updated noise spectrum Ni(f) according to the following formula 7 or 8, or formula 9 or 10 depending on the content of the input voice activity detection result. Note that the subscript i in the following formulas is an ordinal number representing the order in the time series, and represents an order that is commonly applied to all formulas. Also, the subscript j in the following formulas is a variable representing the degree of distance from the ordinal number i in the time series, and is an integer equal to or greater than 0.

When the voice activity detection result input to the noise updating circuit 11 is a noise frame signal, a first updating unit 111 calculates an updated noise spectrum Ni(f) for each frequency f using a signal corresponding to the input amplitude spectrum as an input signal spectrum Yi(f) according to the following formula 7 which is an IIR (Infinite Impulse Response) filter or the following formula 8 which is an FIR (Finite Impulse Response) filter. In the following formulas, Ni-1(f) represents the noise spectrum one frame before the update, and Yi-j(f) represents the input signal spectrum j frames before the update.

Kn in Equation 7 and Equation 8 is a constant that determines the weighting of the noise spectrum Ni-1(f) of the previous frame to the input signal spectrum Yi(f), which is the amplitude spectrum of the frame to be processed (in other words, the current frame or the latest frame) when the frame to be processed is a frame containing only noise components, and is called the "noise pre-update weighting constant Kn."

The noise pre-update weighting constant Kn is not limited to a specific value as long as it is an integer equal to or greater than 0. Specifically, the noise pre-update weighting constant Kn may be set to any value within a range equivalent to the time constant of the IIR filter or the average interval of the FIR filter of about 0.06 to 0.20 seconds (for example, a range of Kn = 5 to 16 at a frame interval of 12.5 ms), and may be set to a value equivalent to the time constant of the IIR filter or the average interval of the FIR filter of about 0.1 seconds (for example, Kn = 8 at a frame interval of 12.5 ms). Note that when the time constant of the IIR filter or the average interval of the FIR filter is set to, for example, 0.1 seconds, the specific value of the constant Kn in Equation 7 differs from the specific value of the constant Kn in Equation 8.

When the voice section detection result input to the noise updating circuit 11 is a voice frame signal, a second updating unit 112 calculates an updated noise spectrum N(f) for each frequency f, using a signal corresponding to the input amplitude spectrum as an input signal spectrum Y(f) in accordance with the following Equation 9 which is an IIR filter or the following Equation 10 which is an FIR filter:

Ks in Equation 9 and Equation 10 is a constant that determines the weighting of the noise spectrum Ni-1(f) of the previous frame to the input signal spectrum Yi(f), which is the amplitude spectrum of the frame being processed (in other words, the current frame or the latest frame) when the frame contains a voice component, and is called the "voice pre-update weighting constant Ks."

The pre-update weight constant Ks during voice is not limited to a specific value, as long as it is a larger integer than the pre-update weight constant Kn during noise, and preferably a sufficiently large integer. Specifically, the pre-update weight constant Ks during voice may be set to any value within a range equivalent to about 1 to 10 seconds of the time constant of the IIR filter or the average interval of the FIR filter (for example, a range of about Ks = 80 to 800 at a frame interval of 12.5 ms), or more specifically, to any value within a range equivalent to about 1 to 6 seconds of the time constant of the IIR filter or the average interval of the FIR filter (for example, a range of about Ks = 80 to 500 at a frame interval of 12.5 ms), and may be set to a value equivalent to about 3.7 seconds of the time constant of the IIR filter or the average interval of the FIR filter (for example, about Ks = 295 at a frame interval of 12.5 ms). Note that when the time constant of the IIR filter or the average interval of the FIR filter is, for example, 3.7 seconds, the specific value of the constant Ks in Equation 9 differs from the specific value of the constant Ks in Equation 10.

By setting the pre-voice weighting constant Ks to a sufficiently large value, voice components are prevented from being significantly reflected in the noise spectrum update due to the non-constant nature of voice, and therefore voice components are not suppressed as noise components.

In addition, when an IIR filter (general form: Ni(f) = ΣAjYi-j(f) + ΣBkNi-k(f); where A and B are coefficients, i is an ordinal number representing the order of the time series, j is an integer of 0 or more representing the degree of distance from ordinal number i in the time series, and k is an integer of 1 or more representing the degree of distance from ordinal number i in the time series) is used as the averaging process for calculating the updated noise spectrum Ni(f), a first updating unit 111 calculates the updated noise spectrum Ni(f) for each frequency f by processing including averaging when the frame to be processed is a frame containing only noise components, using a signal corresponding to the amplitude spectrum of the frame to be processed as the input signal spectrum Yi(f), and a second updating unit 112 calculates the updated noise spectrum Ni(f) for each frequency f by processing including averaging when the frame to be processed is a frame containing voice components. and a second update unit 112 that calculates the updated noise spectrum Ni(f) by processing including averaging, and corresponds to a case where the averaging time of the input signal spectrum Yi(f), the past input signal spectrum Yi-j(f), and the past noise spectrum Ni-k(f) when calculating the updated noise spectrum Ni(f) for a frame containing a voice component is longer than the averaging time of the input signal spectrum Yi(f), the past input signal spectrum Yi-j(f), and the past noise spectrum Ni-k(f) when calculating the updated noise spectrum Ni(f) for a frame containing only noise components.

When using an IIR filter with Aj=0 for j≧1 for the general form of the IIR filter described above, the first update unit 111 calculates an updated noise spectrum Ni(f) for each frequency f by processing including averaging, using a signal corresponding to the amplitude spectrum of the frame to be processed as the input signal spectrum Yi(f) when the frame to be processed is a frame containing only noise components, and the second update unit 112 calculates an updated noise spectrum Ni(f) for each frequency f by processing including averaging when the frame to be processed is a frame containing voice components. This corresponds to the case where the averaging time of the average using the input signal spectrum Yi(f) and the past noise spectrum Ni-k(f) when calculating the updated noise spectrum Ni(f) when the frame contains voice components is longer than the averaging time of the average using the input signal spectrum Yi(f) and the past noise spectrum Ni-k(f) when calculating the updated noise spectrum Ni(f) when the frame contains only noise components.

For the general form of the IIR filter above, if Aj=0 for j≧1, then the case where A0=1/(1+Kn), B1=Kn/(1+Kn) and Bk=0 (k≧2) corresponds to the above formula 7, and the case where A0=1/(1+Ks), B1=Ks/(1+Ks) and Bk=0 (k≧2) corresponds to the above formula 9.

Furthermore, when an FIR filter (general form is Ni(f) = ΣAjYi-j(f); where A is a coefficient, i is an ordinal number representing the order of the time series, and j is an integer equal to or greater than 0 representing the degree of separation from the ordinal number i in the time series; that is, the general form of the IIR filter described above with Bk = 0) is used as the averaging process for calculating the updated noise spectrum Ni(f) for each frequency f, when the frame to be processed is a frame containing only noise components, a first update unit 111 is used which calculates the updated noise spectrum Ni(f) for each frequency f by processing including averaging using a signal corresponding to the amplitude spectrum of the frame to be processed as the input signal spectrum Yi(f); and when the frame to be processed is a frame containing voice components, In some cases, the second update unit 112 calculates the updated noise spectrum Ni(f) for each frequency f by processing including averaging, and the averaging time of the input signal spectrum Yi(f) and the past input signal spectrum Yi-j(f) when calculating the updated noise spectrum Ni(f) for a frame containing a voice component is longer than the averaging time of the input signal spectrum Yi(f) and the past input signal spectrum Yi-j(f) when calculating the updated noise spectrum Ni(f) for a frame containing only noise components.

Note that for the general form of the FIR filter above, when Aj = 1/Kn it corresponds to the above formula 8, and when Aj = 1/Ks it corresponds to the above formula 10.

The noise update circuit 11 outputs a signal corresponding to the updated noise spectrum Ni(f) for each frequency f to the subtraction unit 6 each time it receives an input of a signal corresponding to the amplitude spectrum for each frame and the voice activity detection result. The subtraction unit 6 performs a process of subtracting the signal corresponding to the updated noise spectrum Ni(f) for each frame from the signal corresponding to the amplitude spectrum output from the conversion result output unit 5, using the signal corresponding to the updated noise spectrum Ni(f) output from the noise update circuit 11.

According to the above-mentioned noise update circuit 11, even if the frames to be processed are successive frames containing voice components, it is possible to follow the noise fluctuations and accurately estimate the noise components. Specifically, in a conventional circuit that realizes the spectrum subtraction method, the noise spectrum is updated when the frame to be processed is a frame containing only noise components, but the noise spectrum is not updated when the frame contains voice components. Therefore, if the frames to be processed are successive frames containing voice components, the noise spectrum is not updated, so it is not possible to accurately follow the noise fluctuations, and as a result, it is not possible to accurately estimate the noise components. In contrast, in the above-mentioned noise update circuit 11, the noise spectrum is updated even when the frames to be processed are frames containing voice components, so it is possible to follow the noise fluctuations and accurately estimate the noise components, even if the frames to be processed are successive frames containing voice components.

(Embodiment 2)
FIG. 3 is a functional block diagram showing a schematic configuration of a noise updating circuit 11 according to a second embodiment of the present invention.

This embodiment differs from embodiment 1 in that the noise update circuit 11 has an additional configuration compared to the configuration of embodiment 1 described above, but there are also common configurations and processing contents, and the same configurations and processing contents as embodiment 1 are denoted by the same reference numerals and their description will be omitted.

The noise update circuit 11 according to this embodiment further includes an average calculation unit 113 that calculates an average spectrum level M for the input signal spectrum Yi(f), and a third update unit 114 that determines an updated noise spectrum Ni(f) for a frequency f where Yi(f) ≧ Te×M with respect to the input signal spectrum Yi(f) and the average spectrum level M.

The noise update circuit 11 receives an input of a signal corresponding to the amplitude spectrum for each frame output from the conversion result output unit 5 and branched, and also receives an input of the voice section detection result for each frame output from the voice section detection unit 10. Using the input signal corresponding to the amplitude spectrum, the noise spectrum Ni-1(f) of the previous frame is updated for each frequency f, and the updated noise spectrum Ni(f) is calculated according to any one of the above formulas 7 or 8, 9 or 10, and 12 depending on the contents of the input voice section detection result, etc.

The average value calculation unit 113 calculates an average spectrum level M according to the following formula 11, using the signal corresponding to the amplitude spectrum for each frame output and branched from the conversion result output unit 5 as the input signal spectrum Yi(f). The average spectrum level M is the average value in the frequency axis direction of the amplitude spectrum.

In Equation 11, f represents the frequency in the amplitude spectrum, f1 is the minimum frequency in the amplitude spectrum, f2 is the maximum frequency in the amplitude spectrum, and Fn represents the number of frequencies in the range from the minimum frequency f1 to the maximum frequency f2. In other words, the number of frequencies Fn is n/2+1, where n is the number of sample points included in one frame in the time-frequency conversion by the time-frequency conversion unit 4.

The average calculation unit 113 calculates the average spectrum level M each time it receives an input of a signal corresponding to the amplitude spectrum for each frame.

If the voice section detection result input to the noise update circuit 11 is a noise frame signal, the signal corresponding to the input amplitude spectrum is treated as the input signal spectrum Yi(f), and for each frequency f, the following processing (a) or (b) is performed according to the relationship between the input signal spectrum Yi(f) and the average spectrum level M.

<A> For a frequency f where Yi(f)<Te×M, the first update unit 111 calculates the updated noise spectrum Ni(f) in accordance with Equation 7 or Equation 8 above.

In the relationship between the input signal spectrum Yi(f) and the average spectrum level M, Te is a coefficient for excluding frequency components whose amplitude spectrum is significantly larger than the average spectrum level M from the update of the noise spectrum. The coefficient Te is not limited to a specific value, and is set to an appropriate value, for example, taking into consideration that the frequency components corresponding to the tone signal are excluded from the update of the noise spectrum so that the tone signal is not subtracted by the subtraction unit 6 and is not suppressed as a noise component. Specifically, the coefficient Te may be set to any value in the range of about 1 to 100, and may be set to about 10 in particular.

<A> For a frequency f where Yi(f)≧Te×M, the third update unit 114 determines the updated noise spectrum N i (f) in accordance with the following Equation 12.

In equation 12, Ni-1(f) represents the noise spectrum of the frame before the update.

In addition, if the voice section detection result input to the noise update circuit 11 is a voice frame signal, the signal corresponding to the input amplitude spectrum is treated as the input signal spectrum Yi(f), and for each frequency f, the following processing (c) or (d) is performed according to the relationship between the input signal spectrum Yi(f) and the average spectrum level M.

<c> For a frequency f where Yi(f)<Te×M, the second update unit 112 calculates the updated noise spectrum N i (f) in accordance with Equation 9 or Equation 10 above.

<D> For a frequency f where Yi(f)≧Te×M, the third update unit 114 determines the updated noise spectrum Ni(f) in accordance with Equation 12 above.

In the above process, that is, for frequency components where Yi(f) ≥ Te × M and the amplitude spectrum is significantly greater than the average spectrum level M, the noise spectrum Ni(f) after the update is left as the noise spectrum Ni-1(f) one frame before the update and is excluded from the update.

The noise update circuit 11 outputs a signal corresponding to the updated noise spectrum Ni(f) for each frequency f to the subtraction unit 6 each time it receives an input of a signal corresponding to the amplitude spectrum for each frame and the voice activity detection result. The subtraction unit 6 performs a process of subtracting the signal corresponding to the updated noise spectrum Ni(f) for each frame from the signal corresponding to the amplitude spectrum output from the conversion result output unit 5, using the signal corresponding to the updated noise spectrum Ni(f) output from the noise update circuit 11.

The above-described noise updating circuit 11 makes it possible to avoid suppressing tone signals. Specifically, in conventional circuits that realize the spectral subtraction method, components that are constantly present in the frequency spectrum are judged to be noise and suppressed, so tone signals, which are also stationary, are judged to be noise and are subject to suppression, resulting in the problem that tone signals necessary for the user (e.g., Morse code) are also suppressed. In contrast, the above-described noise updating circuit 11 excludes frequency components whose amplitude spectrum is significantly larger than the average spectrum level from the noise spectrum update, making it possible to avoid suppressing tone signals.

In addition, with respect to the noise updating circuit 11 according to the second embodiment, the manner in which the noise spectrum is updated may be adjusted in accordance with the mode selection in which the user selects whether or not to activate the average value calculation unit 113 and the third update unit 114 depending on the operation at the time, and the manner in which the noise components are suppressed may be adjusted. For example, when the user mainly transmits and receives voice signals while suppressing noise components, the user may select a mode in which the average value calculation unit 113 and the third update unit 114 are not activated to suppress tone signals, whereas when transmitting and receiving Morse codes, the user may select a mode in which the average value calculation unit 113 and the third update unit 114 are activated to prevent tone signals from being suppressed.

Although the embodiment of the present invention has been described above, the specific configuration is not limited to the above embodiment, and even if there are design changes within the scope of the present invention, they are included in the present invention. For example, the above embodiment is described using an example in which the noise updating circuit 11 of the present invention is applied to the noise reduction circuit 1 whose schematic configuration is shown in FIG. 1, but the configuration of the noise reduction circuit to which the present invention can be applied is not limited to the example shown in FIG. 1. Furthermore, the circuits to which the present invention can be applied are not limited to noise reduction circuits. In other words, the present invention can be applied to various circuits in which it is necessary to update the noise spectrum in a time series.

Furthermore, the process including averaging for calculating the updated noise spectrum Ni(f) may be any method that uses coefficients equivalent to the noise pre-update weighting constant Kn and the voice pre-update weighting constant Ks in the above embodiment. In other words, the calculation formula for the updated noise spectrum Ni(f) is not limited to Equation 7 to Equation 10 in the above embodiment.

REFERENCE SIGNS LIST 1 Noise reduction circuit 2 Pre-emphasis circuit 3 Window processing unit 4 Time-frequency conversion unit 5 Conversion result output unit 6 Subtraction unit 7 Synthesis unit 8 Frequency-time conversion unit 9 De-emphasis circuit 10 Voice activity detection unit 11 Noise update circuit 111 First update unit 112 Second update unit 113 Average value calculation unit (Embodiment 2)
114 Third update unit (second embodiment)

Claims (3)

a first updating unit that, when a frame to be processed is a frame containing only noise components, calculates an updated noise spectrum Ni(f) for each frequency f according to the following Equation 1 which is an IIR filter or the following Equation 2 which is an FIR filter, using a signal corresponding to an amplitude spectrum of the frame to be processed as an input signal spectrum Yi(f) (where Ni-1(f): noise spectrum one frame before the update, Yi-j(f): input signal spectrum j frames before the update, Kn: constant determining weighting of Ni-1(f) with respect to Yi(f) when the frame to be processed is a frame containing only noise components, i: ordinal number representing the order in the time series, j: integer equal to or greater than 0 representing the degree of distance from ordinal number i in the time series);
a second update unit that calculates an updated noise spectrum N(f) for each frequency f according to the following Equation 3 which is an IIR filter or the following Equation 4 which is an FIR filter when the frame to be processed is a frame including a voice component (where Ks is a constant that determines the weighting of N(f) with respect to Y(f) when the frame to be processed is a frame including a voice component, Kn<Ks);
having
an average calculation unit that calculates an average spectrum level M for the input signal spectrum Y(f) according to the following formula 5 (where f1 is the minimum frequency in the amplitude spectrum, f2 is the maximum frequency in the amplitude spectrum, and Fn is the number of frequencies in the range from the minimum frequency f1 to the maximum frequency f2);
a third update unit that determines an updated noise spectrum Ni(f) according to the following Equation 6 for a frequency f where Yi(f)≧Te×M with respect to the input signal spectrum Yi(f) and the average spectrum level M (where T is a coefficient, and Ni-1(f) is the noise spectrum one frame before the update);
Further comprising
A noise updating circuit comprising: The constant Ks is set to any value within a range corresponding to the time constant of an IIR filter or the averaging interval of an FIR filter, that is, 1 to 10 seconds.
2. The noise updating circuit according to claim 1 . The coefficient Te is set to any value in the range of 1 to 100.
2. The noise updating circuit according to claim 1 .