JP3202902B2 - Noise suppression device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise suppression device provided in an audio signal input section of, for example, a telephone set for suppressing background noise in voice communication.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, there is one described in the following literature. Reference: U.S. Pat. No. 4,811,404 FIG. 2 is a configuration block diagram showing an example of a conventional noise suppression device described in the reference. In this noise suppression device, for example, an input signal IN from a microphone or the like is input to an input terminal 1, and the input signal IN is input to an analog / digital converter (hereinafter referred to as an ADC).
A / D conversion is performed in 2). Further, the output signal S2 of the ADC 2 is subjected to pre-emphasis processing by the pre-emphasis circuit 3. The pre-emphasis process is a process for enhancing a high frequency component of a signal. Here, the high frequency component is enhanced at a rate of 6 dB / oct. The output signal S3 of the pre-emphasis circuit 3 in which the high frequency components are emphasized is divided by the channel divider 4 into a plurality of frequency bands (hereinafter, referred to as channels) (in the above document, the division is made into 14). The output signal S4 of the channel divider 4 is multiplied by a gain value in a channel gain corrector 5. The output signal S5 of the channel gain corrector 5 multiplied by the gain value
Is output to the channel combiner 6, and the channel combiner 6
At one frequency band. The output signal S6 of the channel coupler 6 is output to the de-emphasis circuit 7, and the high frequency components are attenuated. The rate at which the high frequency components are attenuated is 6 dB / oct, and is canceled by the pre-emphasis processing performed by the pre-emphasis circuit 3, so that the frequency characteristics become flat. The output signal S7 of the de-emphasis circuit 7 is a digital / analog converter (Digital Analog Converter,
D / A conversion is performed by a DAC 8 and an output terminal 9
Is output to, for example, a speaker or a telephone line.

Next, in order to multiply the gain of the output signal S4 of the channel divider 4 divided into each frequency band,
A method of creating a gain value set in the channel gain corrector 5 will be described. The output signal S4 divided into a plurality of channels by the channel divider 4 is output to the channel energy estimator 11
Is input to The channel energy estimator 11 calculates the energy of the input output signal S4. At this time, when the voice metric calculator 12 determines that the input signal (that is, the output signal S4) is an audio signal, the switch 13 is opened, and when it is determined that the signal is noise, the switch 13 is closed. When the switch 13 is closed, the average value of the background noise of the input signal IN is calculated in the following manner. That is, the output signal S11 of the channel energy estimator 11 via the switch 13
Is input to the smoothing filter 14 and the time average of the amplitude is calculated. The length of time for calculating this average is set to, for example, a length of 10 msec. The calculated average value S14 of the amplitude is input to the estimated energy value holder 15, and
It is kept for the sample time (for example, 125 μs). The held value S15 is output to the smoothing filter 14 again for the next averaging operation. Thus, the smoothing filter 14 and the estimated energy holder 15 constitute a low-pass filter for calculating the time average of the amplitude.

On the other hand, an output signal S 15 of the estimated energy value holder 15 is output to a channel S / N (signal level / noise level) estimator 16 and a noise level quantizer 17. The output signal S11 of the channel energy estimator 11 is input to the channel S / N estimator 16. The channel S / N estimator 16 compares the estimated energy value signal average value S15 indicating the average energy of noise previously held in the estimated energy value holder 15 with the channel energy estimated value S11, and calculates the channel S / N. The N estimated value S16 is calculated. This calculation simply performs S / N division. Calculated channel S / N estimate S
16 is output to the channel S / N corrector 18 and the voice metric calculator 12. Voice metric calculator 1
In S2, a signal-to-sound-to-speech probability (ie, certainty that a signal is considered to be speech at a certain S / N) determined by a rule of thumb (not shown) is determined by a rule of thumb. Are referred to as many times as the number of each channel, and the sum of each reference value is calculated. When the total value of the sum exceeds a predetermined threshold value, the input signal IN is determined to be an audio signal, and the switch 13 is opened. On the other hand, when it is determined that the input signal IN is noise, the switch 13 is closed.

[0005] The voice metric calculator 12 calculates the time from when the voice was detected last time. When this time exceeds a predetermined time (for example, one second), this phenomenon is considered as noise. , And the power average of the input signal IN is forcibly recalculated by regarding the input signal IN as noise regardless of the sum of the reference values. The calculation is performed by the smoothing filter 14 and the estimated energy holder 15. Further, the voice metric calculator 12 counts whether the S / N estimation value of each channel exceeds a predetermined threshold. When the counting result is smaller than a predetermined ratio (for example, 4 channels or less) with respect to the total number of channels, the voice metric calculator 12 determines that the input signal IN is a narrow band signal. At this time, the voice metric calculator 12 instructs the channel S / N corrector 18 so that the S / N estimation value S16 corresponding to each channel becomes smaller. The output signal S18 of the channel S / N corrector 18 in which the S / N estimation value S16 has been corrected to a small value is output to the S / N threshold circuit 19.
Then, the S / N threshold circuit 19 refers to the gain table 20 so that the gain table 2 receives a large attenuation.
From 0, a channel gain value S20 is generated.

As described above, the voice metric calculator 12 discriminates three types of signals, that is, background noise, narrow-band noise, and speech, and provides a channel S / N corrector 18 with an S / N of each channel. A signal for correcting the estimated value S16 is output. The output signal S18 of the channel S / N corrector 18 is S
/ N threshold circuit 19. In the S / N threshold circuit 19, when the output signal S18 of the channel S / N corrector 18 has an S / N smaller than a preset threshold, the corrected S / N estimated value S18 is set to the threshold value. I do. The output signal S19 of the S / N threshold circuit 19 is used as a reference input of the gain table 20, and the gain table 20 outputs a gain according to the corrected estimated S / N value S18 to the gain smoothing filter 21. Also, a plurality of gain tables 20 are prepared, and the noise level quantizer 17 determines a gain table to be referred to. Since the selected gain value S20 has a large time variation, the gain value S20 is input to the gain smoothing filter 21 and subjected to a smoothing process. In this smoothing process, the arithmetic mean may be actually obtained for a plurality of samples, or the time constant of the low-pass filter may be appropriately selected to substitute for the averaging operation. The gain S21 output from the gain smoothing filter 21 is output to the channel gain corrector 5, and is multiplied by the gain S21 for each channel of the output signal S4 of the channel divider 4. After the signal subjected to the attenuation processing is output to the channel combiner 6, it is processed in the same manner as described above. As described above, the channel of the output signal S4 of the channel divider 4 which is determined to have many audio signal components is multiplied by a large gain (for example, 1.0 times), and the audio signal component is considered to be small. The sound quality of the input signal IN has been improved by multiplying the channel by a small gain.

[0007]

However, the noise suppressor shown in FIG. 2 has the following problems. (1) When the voice detection period reaches a certain time (for example, one second), the voice metric calculator 12 forcibly updates the estimated background noise. Therefore, this period is actually continued by the voice signal. Also, the contents of the estimated energy value holder 15 that should not be updated may be rewritten, and the quality of the audio signal may be degraded. (2) When an audio signal having a very small noise component is input to the input terminal 1, even if the signal should be passed without any processing, the frequency band is divided and combined. The signal deteriorates. (3) Output signal S from estimated energy value holder 15
Because of the short duration of the estimated energy retention at 14, the noise energy is calculated from the audio signal during the detection time of the voice metric calculator 12, resulting in a correct S / N
Cannot be calculated, and the quality of the audio signal deteriorates.

[0008]

According to the present invention, in order to solve the above-mentioned problems, an ADC for converting an analog audio signal into a digital signal and a high frequency component of an output signal of the ADC are provided to a noise suppression device. A pre-emphasis circuit for emphasizing with a predetermined characteristic, a channel divider for dividing an output signal of the pre-emphasis circuit into channels of a plurality of frequency bands, and an output signal of the channel divider for each of the channels. A channel gain corrector that multiplies the respective attenuation gains, a channel combiner that combines the output signals of the channel gain corrector into a signal of one frequency band, and a high frequency component of the output signal of the channel combiner. A de-emphasis circuit that attenuates the output signal with a characteristic opposite to the characteristic of the pre-emphasis circuit, and an output signal of the de-emphasis circuit. A DAC for converting a digital signal into an analog signal, a channel energy estimator for calculating the energy of the output signal of the channel divider for each channel, and a time average of the output signal of the channel energy estimator for each channel. A smoothing filter, an estimated energy holding unit that holds the output signal of the smoothing filter for each channel corresponding to a time corresponding to the detection time of the audio signal, an output signal of the channel energy estimator, A channel S / N estimator for estimating the S / N of the audio signal for each channel based on an output signal of the estimated energy holder, an output signal of the channel energy estimator and an output of the estimated energy holder A noise level monitor that calculates a noise level of the input signal based on a signal; Based on the output signal of the channel S / N estimator and the output signal of the noise level monitor, a certain period of the audio signal may be an audio period with a small background noise, a noise period with a large background noise, or the background noise. A signal type determiner for determining that the noise fluctuation period fluctuates, and a determination result from the signal type determiner and an output signal of the channel S / N estimator. And a gain table for generating each attenuation gain, and when the determination result of the signal type determination unit is the audio period, an output signal of the ADC is output to the DAC, and the determination result of the signal type determination unit is the noise. Switch means for outputting an output signal of the de-emphasis circuit to the DAC in the case of the period or the noise fluctuation period.

According to the present invention, since the noise suppressor is configured as described above, an analog audio signal is converted into a digital signal by the ADC. Next, the high frequency component of the output signal of the ADC is emphasized with predetermined characteristics in the pre-emphasis circuit. Further, the output signal of the pre-emphasis circuit is divided into a plurality of channels by a channel divider. Here, the energy of the output signal of the channel divider is calculated for each of the channels by a channel energy estimator. Next, a time average of the output signal of the channel energy estimator is calculated for each of the channels by a smoothing filter. Further, the output signal of the smoothing filter is held by the estimated energy holder for each of the channels corresponding to the time corresponding to the detection time of the audio signal. The S / N of the audio signal is estimated for each of the channels by the channel S / N estimator based on the output signal of the channel energy estimator and the output signal of the estimated energy holder. A noise level monitor calculates a noise level of the input signal based on an output signal of the channel energy estimator and an output signal of the estimated energy holder. Further, in the signal type determiner,
Based on the output signal of the channel S / N estimator (ie, the S / N of the audio signal) and the output signal of the noise level monitor, a certain period of the audio signal may be an audio period with a small background noise, A large noise period of the background noise,
Alternatively, it is determined whether the current period is a noise fluctuation period in which the background noise fluctuates.

When the signal type determiner determines that the background noise of the audio signal is small, the output signal of the ADC is directly output to the DAC via a switch. On the other hand, when the signal type determiner determines that the level of the background noise of the audio signal is large, the signal is attenuated to all the channels of the output signal of the channel divider. Further, when the signal type determiner detects that the noise of the audio signal has fluctuated, the output signal of the channel divider is multiplied by each attenuation gain determined for each channel by the channel gain corrector. You. Each output signal of the channel gain corrector is combined into a signal of one frequency band by a channel combiner. The high frequency component of the output signal of the channel combiner is attenuated by a de-emphasis circuit with a characteristic opposite to that of the pre-emphasis circuit. Further, the output signal of the de-emphasis circuit is converted from a digital signal to an analog signal by the DAC. Therefore, in the noise suppression device of the present invention, even if the audio signal is continuous for a certain period of time or more, the audio signal is not erroneously recognized as background noise and unnecessary attenuation processing is not performed. Communication is performed with good voice quality without signal loss. Therefore, the above problem can be solved.

[0011]

FIG. 1 is a block diagram showing the configuration of a noise suppression apparatus according to an embodiment of the present invention. This noise suppression device has an input terminal 31 for inputting an input signal IN. The input terminal 31 is connected to the input terminal of the ADC 32. A first output terminal of the ADC 32 is connected to an input terminal of the pre-emphasis circuit 33. An output terminal of the pre-emphasis circuit 33 is connected to an input terminal of the channel divider 34. The output terminal of the channel divider 34 is connected to the input terminal of the channel gain corrector 35. An output terminal of the channel gain corrector 35 is connected to an input terminal of the channel combiner 36. The output terminal of the channel combiner 36 is connected to the input terminal of the de-emphasis circuit 37. De-emphasis circuit 37
Is connected to the first terminal of the DAC 39 via the switch 38.
Is connected to the input terminal of The output terminal of the DAC 39 is connected to the output terminal 40. On the other hand, the output terminal of the channel divider 34 is connected to a channel energy estimator 41.
Is connected to the input terminal of An output terminal of the channel energy estimator 41 is connected to a first input terminal of the smoothing filter 43 via a switch 42. The output terminal of the smoothing filter 43 is connected to the input terminal of the estimated energy holder 44. A first output terminal of the estimated energy holder 44 is connected to a second input terminal of the smoothing filter 43. A second output terminal of the estimated energy holder 44 is connected to a first input terminal of the channel S / N estimator 45. An output terminal of the channel energy estimator 41 is connected to a second input terminal of the channel S / N estimator 45. Channel S / N estimator 45
Is connected to the first input terminal of the signal type determiner 46. A first output terminal of the signal type determiner 46 is connected to a first input terminal of the noise level monitor 47 and to an input terminal of the switch controller 48. A third output terminal of the estimated energy holder 44 is connected to a second input terminal of the noise level monitor 47. An output terminal of the noise level monitor 47 is connected to a second input terminal of the signal type determiner 46.

On the other hand, the on / off state of the switch 42 is controlled by a signal type determiner 46. A second output terminal of the ADC 32 is connected to a second input terminal of the DAC 39 via a switch 49. This switch 49 and the switch 38
The on / off state is controlled by the switch controller 48. The output terminal of the channel S / N estimator 45 is connected to the input terminal of the S / N threshold circuit 50. An output terminal of the S / N threshold circuit 50 is connected to a first input terminal of the gain table 51.
A second output terminal of the signal type determiner 46 is connected to a second input terminal of the gain table 51. An output terminal of the gain table 51 is connected to an input terminal of the gain smoothing filter 52. An output terminal of the gain smoothing filter 52 is connected to a control input terminal of the channel gain corrector 35. Next, the operation of FIG. 1 will be described.

The input signal IN input from the input terminal 31
Is subjected to A / D conversion by the ADC 32, and an output signal S 32 of the ADC 32 is input to the pre-emphasis circuit 33.
The pre-emphasis circuit 33 emphasizes high frequency components of the output signal S32. The emphasis rate is generally 6 dB / oct, but may be appropriately determined according to the application. The output signal S33 of the pre-emphasis circuit 33 is
And is divided into a plurality of channels. In the present embodiment, the plurality of channels are, for example, 14 to 15 channels. The number of divisions may be any number as long as it is appropriately selected according to the form to which the system is applied. The output signal S34 of the channel divider 34 is multiplied by an attenuation gain for each channel in the channel gain corrector 35, and is appropriately attenuated to become an output signal S35. The output signal S35 is input to the channel combiner 36 and is combined into one channel. The output signal S36 of the channel combiner 36 is input to the de-emphasis circuit 37, and the high frequency components are attenuated at a rate of 6 dB / oct. The output signal S37 of the de-emphasis circuit 37 is output to the DAC 39 via the switch 38,
After the D / A conversion, it is output to the output terminal 40.

On the other hand, the output signal S3 of the channel divider 34
4 is input to the channel energy estimator 41. Output signal S3 input to channel energy estimator 41
4 is represented by X (K, t) where K; channel number t; time, and the channel energy estimator 41 calculates an absolute value or a square value. For selection of either the absolute value or the squared value, an appropriate value may be selected when used. Hereinafter, as an example, the output signal S34, that is, X (K,
The absolute value of t) | X (K, t) | where K; 1, 2,..., NN; the number of channels t;

First, the switch 49 is turned off and the switch 38 is turned on. When the result of the determination by the signal type determiner 46 is no sound, the switch 42 is always turned on. In other cases, the switch 42 is turned off. In the initial state, the switch 42 is turned on in response to a command from the signal type determiner 46. The output signal S41 of the channel energy estimator 41 is output to the smoothing filter 43 for each channel. In the smoothing filter 43, an operation is performed using the following equation (1). H (K, t) = (1−C1) × | X (K, t) | + C1 × H (K, t−1) (1) where C1; the time constant of the averaging operation H (K, t-1): Time t at the estimated energy holder 44
−1, that is, the estimated energy value held in the past one sample period. The estimated energy retainer 44 retains values for J (J; integer> 0) samples of H (K, t−1) to H (K, t−J). The constant J may be appropriately determined in the range of J> 0 depending on the application.

Here, the initial value at time t = 0 is H (K, 0) = 0. The time constant C1 of the averaging operation is, for example, 1
It is set to correspond to seconds. That is, the smoothing filter 4
3 and the estimated energy holding unit 44 have a function of smoothing the channel energy estimated value S41 having a large time variation and holding the value. An output signal S44 of the estimated energy holder 44 is output from a channel S / N estimator 45.
Is input to On the other hand, the output signal | X (K, t) | of the channel energy estimator 41 is also input to the channel S / N estimator 45. Channel S
The / N estimator 45 receives the output signal H (K, t) of the estimated energy holder 44 as the noise energy N (K, t) and the output signal | X (K, t) | of the channel energy estimator 41. Assuming the current energy of the signal IN, an S / N estimate of the input signal IN is calculated for each channel. In the present embodiment, it is assumed that the channel S / N estimator 45 calculates the S / N estimated value S / N (K, t) (dB) using the following equation (2). S / N (K, t) = 20 · log (| X (K, t) | / N (K, t)) (2) where K; 1,..., NN; t; time Depending on the system to be used, | X (K, t) | / N (K, t) in the above equation (2) may be directly calculated and used for the subsequent processing.

The estimated S / N (K,
t) is input to the signal type determiner 46. On the other hand, the noise level monitor 47 has a channel energy estimator 4
1 (ie, | X (K, t) |) and the output signal S44 of the estimated energy holder 44 (ie, H
(K, t)). Noise level monitor 4
7 indicates that if the magnitude of the output signal S41 | X (K, t) | of the channel energy estimator 41 is a continuously small value, for example, smaller than a preset threshold, there is a possibility that the background noise is small. And notifies the signal type determiner 46 that the background noise is small. The signal type determiner 46 receives this signal and turns on the switch 42. as a result,
As shown in equation (1), in the smoothing filter 43,
The level average or the energy average value for a certain period, for example, one second, is calculated. If the calculated energy average value (that is, the smoothed value) is equal to or less than a preset threshold,
The signal type determination unit 46 determines that the background noise is small, and outputs a control signal S46 to the switch controller 48. At this time, the output signal S48 of the switch controller 48
Is a signal for controlling the switch 49 to be turned on and the switch 38 to be turned off. Also, the switch 42 remains on. At this time, the input signal IN
Are output directly to the output terminal 40 via the ADC 32, the switch 49, and the DAC 39 in order. For this reason, the audio signal does not suffer from deterioration in sound quality due to channel division and channel coupling.

Next, the case where the background noise is relatively large will be described. When the magnitude of the output signal | X (K, t) | of the channel energy estimator 41 is continuously larger than a preset threshold, a predetermined period from the initial time (that is, t = 0), for example, 1
The switch 42 is turned on for a second, and the background noise energy level N (K, t) is calculated using the smoothing filter 43 and the estimated energy holder 44. The calculated background noise energy level N (K, t) is output to the noise level monitor 47, and the background noise energy level N
If (K, t) is larger than the threshold, the switch control circuit 4
8 turns the switch 49 off and the switch 38 on. On the other hand, the output signal S44 of the estimated energy holder 44 and the output signal S41 of the channel energy estimator 41 are input to the channel S / N estimator 45. In the channel S / N estimator 45, the calculated background noise energy level N (K, t) and the output signal S41 of the channel energy estimator 41 (that is, | X (K, t) |)
From the channel S / N estimated value SN shown in the following equation (3)
(K, t) is calculated and input to the signal type determiner 46. SN (K, t) = | X (K, t) | / N (K, t) (3) where K; channel number 1,..., Nt; , For a preset period ΔT (eg, 1 second), the channel S / N estimation value S /
The change amount ΔS / N of N (K, t) is obtained.

FIG. 3 shows the estimation S of the voice section in each channel.
FIG. 10 is a diagram illustrating an example of a change in / N, in which the vertical axis represents estimated S / N and the horizontal axis represents time. As shown in this figure, an audio signal is generally a signal that fluctuates with time greater than background noise. Therefore, if the variation ΔS / N of the estimated channel S / N value S / N (K, t) is larger than a preset threshold, the signal type determination unit 46 determines that this signal period is an audio signal period. Switch 42 is opened to stop updating the background noise energy. At this time, the estimated energy holding unit 44 outputs the estimated energy value H (K, t−I) to the channel S / N estimator 45, where I: the output signal S44 satisfying the integer I <J is the background noise N (K, t). The constant J is equal to the estimated energy holder 44 described above.
Is an integer representing the limit number of samples that can be held. FIG.
Is a diagram showing an example of an estimated S / N change in a noise fluctuation section in each channel, where the estimated S / N is plotted on the vertical axis and time is plotted on the horizontal axis.

When the background noise level itself fluctuates, as shown in FIG. 4, the channel S / S
Since the amount of change ΔS / N of the estimated N value becomes smaller, when ΔS / N becomes smaller than a preset threshold value, the signal type determiner 46 considers that the background noise has changed during this observation period, 42 is turned on. The output signal S41 of the channel energy estimator 41 is output to the smoothing filter 43 through the switch 42, and the smoothing filter 43 and the estimated energy holder 44
A time average value N (K, t) serving as a background noise estimated value is calculated, and the estimated energy value H (K, t-1) held in the estimated energy holding unit 44 is converted to the background noise N (K, t).
Output to the channel S / N estimator 45 as t). In the channel S / N estimator 45, the channel S / N
N is calculated as described above, and the signal type
And to the S / N threshold circuit 50. S / N threshold circuit 5
In the case of 0, an instruction is given to a channel having an S / N value larger than a preset threshold value so that the gain table 51 can be referred to. That is, for a channel having an estimated S / N value equal to or larger than the threshold value, the gain table 51 is referred to so that an attenuation gain corresponding to the estimated S / N value is given to each channel. For a channel having an S / N (for example, 3 dB) that is too small to refer to the gain table, the maximum attenuation amount (for example, 1
0 dB).

The attenuation gain G (K, t) generated from the gain table 51 is output to a gain smoothing filter 52. The gain smoothing filter 52 calculates a smoothing gain Gs (K, t) using the following equation (4). Gs (K, t) = (1−Cg) × (Gs (K, t)) + Cg × G (K, t) (4) where K; channel number 1,..., Nt; Time Cg; smoothing constant Here, the smoothing constant Cg may be appropriately selected under the condition of Cg <1. Output signal S52 of gain smoothing filter 52
(That is, Gs (K, t)) is the channel gain corrector 35
And the output signal S34 of the channel divider 34 receives an appropriate amount of attenuation for each channel. That is, the signal is hardly attenuated for the channel containing the audio signal component strongly,
A channel having a small amount of audio signal components or a majority of noise components is greatly attenuated and output to the channel coupler 36.

In the channel combiner 36, all the channels divided by the channel divider 34 are combined into one band signal. However, each channel is a signal whose sound quality has been improved by attenuating noise components. Channel coupler 3
6 is output to the de-emphasis circuit 37 to attenuate high frequency components. The rate of attenuation is 6 dB / oct in the present embodiment. The output signal S37 of the de-emphasis circuit 37 is output to the DAC 39,
The DAC 39 performs D / A conversion on the audio signal whose sound quality has been improved, and outputs the signal to the output terminal 40. As described above, in the present embodiment, when the signal type determination unit 46 determines that the background noise of the audio signal IN is small, the output signal of the ADC 32 is directly output to the DAC 39 via the switch 49. The audio signal IN does not undergo unnecessary band division, combining, and emphasis processing. On the other hand, when the signal type determiner 46 determines that the level of the background noise of the audio signal IN is large, attenuation is applied to the channel whose estimated S / N calculated by the channel S / N estimator 45 is small. or,
When the signal type determiner 46 detects that the noise of the audio signal IN has changed, according to the changed noise level,
The output signal S34 is attenuated for the channel with increased noise. For this reason, the noise suppression device of FIG. 1 does not perform unnecessary attenuation processing by erroneously recognizing the audio signal IN as background noise even if the audio signal IN is continuous for a predetermined time or more. The communication is performed with good voice quality without loss of the voice signal to be transmitted.

The present invention is not limited to the above embodiment,
Various modifications are possible. For example, there are the following modifications. (A) The emphasis and attenuation ratios of the high frequency components of the pre-emphasis circuit 33 and the de-emphasis circuit 37 are
Pre-emphasis circuit 33 and de-emphasis circuit 3
Any ratio may be used as long as it is set so that the effect of 7 is offset. (B) The switch 38 and the switch 49 are connected to the ADC 32
Or an output signal of the de-emphasis circuit 37 to select and output the selected signal to the DAC 39.

[0024]

As described above in detail, according to the present invention, when the signal type determiner determines that the background noise of the audio signal is small, the output signal of the ADC is directly output to the DAC via the switch means. Therefore, the audio signal is not subjected to unnecessary band division, combination, and emphasis processing. On the other hand, when the signal type determiner determines that the background noise level of the audio signal is large, attenuation is applied to the channel indicating the low S / N estimation value of the output signal of the channel divider. When the signal type determiner detects that the noise of the audio signal has changed, attenuation is applied to the channel in which the noise of the output signal of the channel divider has increased according to the changed noise level. Therefore, the noise suppression device of the present invention does not perform unnecessary attenuation processing by erroneously recognizing the audio signal as background noise even if the audio signal is continuous for a certain period of time or more. Communication can be performed with good voice quality without signal loss.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a noise suppression device according to an embodiment of the present invention.

FIG. 2 is a configuration block diagram of a conventional noise suppression device.

FIG. 3 is a diagram illustrating a change example of estimated S / N of a voice section in each channel.

FIG. 4 is a diagram illustrating an example of a change in estimated S / N of a noise fluctuation section in each channel.

[Explanation of symbols]

32 ADC 33 pre-emphasis circuit 34 channel divider 35 channel gain corrector 36 channel combiner 37 de-emphasis circuit 38,49 switch (switch means) 39 DAC 41 channel energy estimator 43 smoothing filter 44 estimated energy holder 45 channel S / N
Estimator 46 Signal type determiner 47 Noise level monitor 51 Gain table 52 Gain smoothing filter

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04B 15/00 H03M 1/08 H04B 1/62

Claims (1)

(57) [Claims] An analog / digital converter for converting an analog audio signal into a digital signal; a pre-emphasis circuit for emphasizing a high frequency component of an output signal of the analog / digital converter with predetermined characteristics; A channel divider that divides an output signal of the emphasis circuit into channels of a plurality of frequency bands, and a channel gain corrector that multiplies an output signal of the channel divider by each attenuation gain obtained for each channel. A channel combiner that combines each output signal of the channel gain corrector into a signal of one frequency band; and attenuates a high frequency component of the output signal of the channel combiner with a characteristic opposite to that of the pre-emphasis circuit. A de-emphasis circuit for converting an output signal of the de-emphasis circuit from a digital signal to an analog signal A digital / analog converter for converting the energy of the output signal of the channel divider into a channel energy estimator for each of the channels; and a time average of the output signal of the channel energy estimator for each of the channels. A smoothing filter to be calculated; an estimated energy holding unit that holds an output signal of the smoothing filter for each channel corresponding to a time corresponding to a detection time of the audio signal; an output signal of the channel energy estimator; A channel S / N estimator for estimating the S / N of the audio signal for each channel based on an output signal of the estimated energy holder; an output signal of the channel energy estimator and an output signal of the estimated energy holder A noise level monitor that calculates a noise level of the input signal based on: / N based on the output signal of the / N estimator and the output signal of the noise level monitor, a certain period of the voice signal is a voice period with a small background noise, a noise period with a large background noise, or a fluctuation of the background noise. A signal type determiner for determining that the noise fluctuation period is to occur, a determination result from the signal type determiner, and the channel S
/ N based on the output signal of the / N estimator, a gain table for generating an appropriate attenuation gain for each channel of the channel divider, and when the determination result of the signal type determiner is the voice period,
An output signal of the analog / digital converter is output to the digital / analog converter. If the determination result of the signal type determination unit is the noise period or the noise fluctuation period, the output signal of the de-emphasis circuit is converted to the digital signal. / A switch means for outputting to an analog converter.