JP2008048281A - Noise reduction apparatus, noise reduction method and noise reduction program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To eliminate random noise that is one of blocking factors for high sound quality recording and reproduction of a video camera or the like including a plurality of microphone systems, and to also improve the problem that a sound field becomes monaural. <P>SOLUTION: A noise reduction apparatus includes: an input means for inputting a plurality of acoustic signals from a plurality of acoustic channels; a noise extracting means; an arithmetic means; a plurality of first level detecting means each for detecting a signal level, during a prescribed period of time, of a signal from the noise extracting means; a second level detecting means for detecting a signal level, during a prescribed period of time, of a signal from the arithmetic means; a selection means for selecting, for each prescribed period of time, a signal having a minimum level value among level values detected from the first and the second level detecting means; a band limiting means for performing a band limitation on a signal from the selection means; and a plurality of composing means for composing, for each acoustic channel, a signal from the band limiting means and a signal other than the extracted level in the noise extracting means, wherein output of each composing means is defined as each acoustic channel output signal. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a noise reduction device, a noise reduction method, and a noise reduction program for reducing noise mixed in an audio signal from a microphone or the like.

  In general, in devices with built-in microphones such as video cameras and digital cameras, thermal noise and quantum generated from microphones, analog elements such as preamplifiers and resistors, and semiconductor elements such as operational amplifiers and analog-digital converters (ADCs). There is a problem that random noise (also called white noise, noise whose level is uniform at all frequencies and whose phase is random) is mixed into the audio signal.

  Although these random noises are always generated, especially when shooting in a quiet place, the sensitivity of the microphone is increased by an internal AGC (Automatic Gain Control) circuit, so this noise is often noticeable. Furthermore, in recent video cameras, a plurality of microphones are mounted, and in order to perform sound field processing on the output from each microphone, random noise generated from each microphone system circuit is emphasized by this sound field processing. This problem has become more prominent.

  First, FIG. 19 shows a stereo sound field processing apparatus in a general video camera disclosed in Patent Document 1. 19, the right channel acoustic signal (Rch signal) and the left channel acoustic signal (Lch signal) are input from the microphones 1 and 2, respectively, and the signal level is optimized by the preamplifiers (AMP) 3 and 4, The signal is input to a delay terminal (DL) 5 which is composed of a positive terminal of the adder 9 and generally a low pass filter (LPF). Similarly, the Lch signal is input to the + terminal of the adder 10 and the delay device (DL) 6, and the low frequency components of the signal band delayed by the DL 5 and 6 are leveled by the attenuators (ATT) 7 and 8, respectively. Is input to the negative terminals of the adders 10 and 9.

  In the adders 9 and 10, matrix processing is performed by subtracting the low-frequency component signals from the channels, that is, signals below the cutoff frequency of the LPF set by the DL5 and DL6, and stereo processing is performed in the low-frequency band. The sound field is reproduced. Further, the outputs of the adders 9 and 10 are adjusted in frequency characteristics by equalizers (EQ) 11 and 12 constituted by filter circuits, and output from the output terminals 13 and 14 as Rch signals and Lch signals.

  Further, FIG. 20 shows stereo directivity characteristics generated by the stereo sound field processing apparatus of FIG. First, in the stereo sound field processing apparatus of FIG. 19, since the microphones 1 and 2 generally use non-directional microphones having no directivity, the signals output from the microphones 1 and 2 are the microphones 1 and 2. 2 is generated, and the distance difference is sufficiently smaller than the distance to the input sound source. Therefore, the Lch signal (solid line) and the Rch signal (broken line) shown in FIG. ) Shows a characteristic close to monaural.

  Then, after the arithmetic processing of the stereo sound field processing device of FIG. 19, a phase difference occurs as shown in FIG. 20B, and the directivity pattern is divided into an Lch signal (solid line) and an Rch signal (broken line) in the left-right direction. Stereo characteristics. Thus, the process of the stereo sound field processing apparatus of FIG. 19 can be said to be a process of enlarging a slight phase difference.

  By the way, since the random noise described above is generated with a random phase difference between the Lch signal and the Rch signal, when the stereo noise field processing device shown in FIG. When added in the same phase, the level deteriorates by 6 dB, and the gain of the noise band is further increased by EQs 11 and 12 in the subsequent stages. And since noise has a stereo characteristic as shown to FIG. 20B, it spreads to the whole sound field, and even if the noise to generate | occur | produce is a very minute level, the malfunction which will be emphasized will arise.

  The noise described above is also referred to as white noise, and is present in a wide band over the entire frequency band of the voice and at a minute level as compared with the voice signal, and therefore cannot be uniformly removed by a filter, for example.

  Furthermore, with the recent increase in screen size and HD of television receiver TVs, the sound reproduction system has also been improved in sound quality, and there are increasing opportunities to perform 5.1ch surround sound field reproduction. There is also a strong demand for higher sound quality.

Therefore, Patent Document 2 discloses a technique in which random noise is reduced by an averaging process.
Japanese Patent No. 2946638 JP 2006-50067 A

However, the one disclosed in Patent Document 2 has a disadvantage that the sound field is made monaural because the output signals of a plurality of microphones are subjected to an averaging process to become a monaural signal.
In view of such a point, the present invention eliminates random noise, which is one of the obstruction factors for high-quality recording and reproduction of a video camera or the like having a plurality of microphones, and makes the sound field monophonic. Also aimed at improving.

  The noise reduction device of the present invention includes input means for inputting a plurality of sound signals from a plurality of sound channels, a plurality of noise extraction means for extracting a level region including a noise component in the sound signals, and noise from the noise extraction means. An arithmetic means for calculating an addition average of the signals, a plurality of first level detecting means for detecting a signal level in a predetermined period of the signal from the noise extracting means, and a signal level in the predetermined period of the signal from the arithmetic means are detected. Second level detecting means for selecting, and a selecting means for selecting a signal having a level value with the smallest level value detected from the first level detecting means and the second level detecting means for each predetermined period. A band limiting unit for limiting the band of the signal from the selection unit, a signal from the band limiting unit, and a noise extracting unit. And a signal other than the extraction level and a plurality of combining means for combining for each sound channel respectively, the output of each combining means is obtained so as to each acoustic channel output signal.

  The noise reduction apparatus of the present invention includes an input unit that inputs a plurality of acoustic signals from a plurality of acoustic channels, a plurality of noise extraction units that extract a level region including a noise component in the acoustic signal, and a signal from the noise extraction unit. Calculating means for calculating an average of the signals, a plurality of first level detecting means for detecting a signal level in a predetermined period of the signal from the noise extracting means, and detecting a signal level in the predetermined period of the signal from the calculating means The second level detecting means, the level value from the first level detecting means and the level value from the second level detecting means are set to the level value of the smaller level for each predetermined period in each acoustic channel. A plurality of selection means for selecting a signal, a plurality of band limitation means for limiting the band of the signal from the selection means, and the band limitation means A plurality of synthesis means for synthesizing the signal and a signal other than the extraction level in the noise extraction means for each acoustic channel, and the output of each synthesis means is used as each acoustic channel output signal. is there.

  The noise reduction device of the present invention includes an input unit that inputs a plurality of acoustic signals from a plurality of acoustic channels, a band extraction unit that extracts a predetermined band from the acoustic signals, and a level that includes a noise component in the signal from the band extraction unit. A plurality of noise extracting means for extracting a region, a plurality of calculating means for calculating an average of signals from the noise extracting means for each predetermined band, and a signal level in a predetermined period of the signal from the noise extracting means are detected. A plurality of first level detecting means, a second level detecting means for detecting a signal level in a predetermined period of the signal from the calculating means, a level value from the first level detecting means and the second level detecting means. A selection means for selecting a signal having the lowest level value from the level detection means for each predetermined period of the predetermined band; and A band limiting unit that limits the band of the signal from the selection unit, a band synthesizing unit that performs band synthesis for each acoustic channel, a signal from the band limiting unit, a band signal other than the extraction band in the band extracting unit, A plurality of synthesizing units for synthesizing the signal from the band synthesizing unit and the signal other than the extraction level in the noise extracting unit for each acoustic channel, and the output of the synthesizing unit is used as each acoustic channel output signal; It is a thing.

  According to the present invention, the target noise is reduced by adding and averaging because there is no correlation between the channels and the phase is random noise. However, in the present invention, in addition to this, the minimum of the input signal is reduced. Since the phase difference component can be strongly removed by performing the selection process, the reduction effect can be improved even with a small number of microphones.

  In addition, according to the present invention, since only the noise level region below the predetermined level is processed, it can be reduced with little influence on the audio signal, and after the noise addition averaging process, random noise is detected in each channel. Since the signal is a monaural signal, for example, even if sound field calculation processing is performed from a plurality of channels in the subsequent stage, the noise level does not deteriorate.

  According to the present invention, noise can be reduced while leaving a sound field feeling (independence) by performing a minimum selection process on the noise addition average processing output and each acoustic channel output.

  According to the present invention, by performing band selection, performing minimum selection processing for each divided band, and performing band synthesis again at the final stage, it is possible to reduce noise optimally for each band, and to provide an acoustic signal. The influence is small and the reproducibility of the acoustic signal is improved.

  Hereinafter, an example of the best mode for carrying out the noise reduction device, the noise reduction method, and the noise reduction program of the present invention will be described with reference to the drawings.

  FIG. 1 shows a noise reduction apparatus according to this example. As will be described with reference to FIG. 1, an Rch signal from an Rch input terminal 30 is a low level signal (Low) whose input signal is equal to or lower than a predetermined level by a level discriminating means 32 described later. Output) and other high level signals (High output). Similarly, the Lch signal from the Lch input terminal 31 is also separated into a low level signal and other high level signals by the level discrimination means 33. .

  Further, the high level signal of each of the level discriminating means 32 and 33 is supplied to one + side terminal of the adders 39 and 40, and the low level signal of each of the level discriminating means 32 and 33 is supplied to the + terminal of the adder 34. The level value detection / determination means 36 and the first fixed contact 37a and the third fixed contact 37c of the switching switch 37 are supplied. Then, from the adder 34, the (L + R) / 2 composite signal is converted into the (L + R) ch signal via the ½ calculator 35 in the same manner as the level value detection / determination means 36 and the second fixed switch 37. It is supplied to the contact 37b.

  Here, the level discriminating means 32 and 33 will be described with reference to FIGS. Since the level discriminating means 32 and 33 are configured in the same manner, they will be described as arbitrary channels in order to avoid duplication. An input signal from the input terminal 20 is supplied to the detecting means 22 via the changeover switch 25 and the absolute value converting means 21 for level detection. The operation at this time will be described with reference to FIG. The input signal having an arbitrary waveform as shown in FIG. 3A is converted into an absolute value by the absolute value converting means 21 as shown by the solid line in FIG. 3B, and further detected by the detecting means 22 as indicated by the short broken line in FIG. 3B. Level detected.

  The output signal of the detection means 22 is supplied to the level comparison means 23 and is compared with a reference level (extraction level) separately input from the reference level input terminal 24 by the level comparison means 23 of FIG. 2, and the detection level is shown in FIG. 3B. A period smaller than the reference level indicated by the long broken line is determined as a low level signal period, and the other period is determined as a high level signal period, and this determination output is supplied to the changeover switch 25 to move the movable contact 25a. Control. Then, the movable contact 25a of the changeover switch 25 is switched based on this determination result. When the input signal is in the high level signal period, it is connected to the fixed contact 25b on one side, and is output to the output terminal 26. In this case, it is connected to the other fixed contact 25 c and outputted to the output terminal 27.

  Further, the noise described above is present in a large amount in the low level signal, and each of the low level signals of the Rch signal, the (L + R) ch signal, and the Lch signal including these noises has a level value detection / configuration configured as shown in FIG. The level detection / determination process is performed by being supplied to the determination unit 36. Referring to FIG. 4, a low level signal of the Rch signal is input from the Rch input terminal 50, a low level signal of the (L + R) ch signal is input from the (L + R) ch input terminal 51, and the Lch signal is input from the Lch input terminal 52. Are detected by the level detection circuits 56, 57, and 58 via the absolute value processing circuits 53, 54, and 55, respectively, and the level value is determined based on the detected level value. The level is determined by the means 59, and the determination output is output from the output terminal 60.

  Here, the operation of the level value determining means 59 will be described with reference to the flowchart of FIG. First, an Rch level value, an Lch level value, and an (L + R) ch level value ((L + R) / 2 composite signal level value) are input. In step S1, (L + R) ch level value ≦ Lch level value is determined. If yes, (L + R) ch level value ≦ Rch level value is determined in step S2, and if yes, the (L + R) ch level value is set as a determination output (step S3). If NO in step S1, Rch level value ≦ Lch level value is determined in step S4. If YES, the Rch level value is set as a determination output (step S5), and no in step S4. For example, the Lch level value is set as a determination output (step S6), and each set determination is output.

  Therefore, since the determination output always operates so that the signal having the lowest level is selected, the movable contact 37d of the changeover switch 37 in FIG. 1 is switched according to this determination output, and the lowest level is obtained within the predetermined period. A signal is supplied to the other + side terminal of each of the adders 39 and 40 via the LPF 38, and is combined with each high level signal and output as an Rch signal and an Lch signal from the Rch output terminal 41 and the Lch output terminal 42. Is done.

  Here, the operation of FIG. 1 will be described with reference to FIG. First, the low level signal from the level discriminating means 33 is the Lch signal in FIG. 6A, and the low level signal from the level discriminating means 32 is the Rch signal in FIG. 6B. Further, the adder 34 and the 1/2 calculator 35 generate the (L + R) / 2 composite signal in FIG. 6C. Further, the level value detection / determination means 36 performs level comparison for each input signal as shown in FIG. 4, supplies this determination output to the changeover switch 37 to control the movable contact 37d, and the minimum signal thereof. Is selected and output as shown by the thick line in FIG. 6D. Further, when LPF 38 is passed to suppress the harmonic component at the time of switching, the output is output as shown in FIG. 6E.

  As described above, since the target noise of this example has phase randomness between channels, even if the signals between the channels are simply added and averaged, the noise level is as shown in the waveforms of the time scales t8 to t13 in FIG. 6C. Although the level is suppressed in the portion where the phase difference is large, the level of FIG. 6E which is the signal of this example is reduced compared to this, and a component having no correlation between channels such as random noise is strongly strengthened. There is a feature that the components that are removed and have a strong correlation between channels such as an audio signal do not affect much.

  Next, FIG. 7 shows an example 2 of the level discriminating means. 7 will be described. Parts corresponding to Example 1 of the level discriminating means in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted. 7 is obtained by changing the changeover switch 25 of FIG. 2 to the crossfade switching means 75, and gradually changes over a predetermined time at ON / OFF as shown in the crossfade switching characteristic. Is set to about several milliseconds. As a result, it is possible to suppress the occurrence of instantaneous noise at the time of a high level signal and at the time of a low level signal.

  Further, Example 3 of the level discriminating means in FIG. 8 will be described with reference to FIGS. 9 and 10. In this example, including the above-described level discriminating examples 1 and 2, the human auditory masking phenomenon is used in this example. , Level discrimination. This is because human hearing is unaware of the presence of a small sound, such as the aforementioned noise, that is behind a relatively loud sound so that, for example, in a loud noise, it is difficult to hear a human voice. Therefore, the input sound level is detected, and when the detected level is high, level discrimination is not performed.

  The example 3 of the level discriminating means in FIG. 8 has the same concept, and the input signal from the input terminal 200 is supplied to the + side terminal of the adder 202, and the adder 202 is connected via the low level extraction circuit 201. The signal is supplied to the negative terminal and is output as a low level signal from the low output terminal 204, and the subtraction result of the adder 202 is output as a high level signal from the high output terminal 203. The low level extraction circuit 201 will be described with reference to FIG.

  An input signal from the input terminal 210 is supplied to one fixed contact 215a of the changeover switch 215 and is also supplied to the detection means 212 via the absolute value conversion means 211 for level detection. Detected. Then, the level comparison means 213 compares the reference level (extraction level) separately input from the reference level input terminal 214, and determines the period in which the detection level is smaller than the reference level shown by the long broken line in FIG. 3B as the low level signal period. The other period is determined as the high level signal period, and the determination output is supplied to the changeover switch 215 to control the movable contact 215c. Based on the determination result, the changeover switch 215 connects the movable contact 215c to one fixed contact 215a in the case of the low level signal period, outputs the input signal to the output terminal 216, and in the case of the high level signal period. The movable contact 215c is connected to the other fixed contact 215b, and GND (no signal) is output. An example at this time will be described with reference to FIG.

  FIG. 10A shows the input signal of the input terminal 210 in FIG. 9, and the reference level indicated by the broken line is set so as to pass noise at a minute level compared to the audio signal, and the audio signal above this reference level is set. As shown in FIG. 10B, the signal gate period is replaced with no signal.

  Therefore, in FIG. 9, the high level signal can be output as it is during the high level signal period with a high audio level to avoid the influence on the audio, and the low level signal can be output during the low level signal period. The high level signal obtained by subtracting the low level signal component from the high level signal can be output.

  FIGS. 11, 12, 13, 14 and 15 show other examples of the best mode for carrying out the present invention. FIG. 11, FIG. 12, FIG. 13, FIG. 14 and FIG. Parts corresponding to 1 are denoted by the same reference numerals or subscripts, and detailed description thereof is omitted.

  11 will be described. In this example of FIG. 11, the level discriminating means 32 and 33 of FIG. 1 are replaced with a level detecting means 83 which is a level detecting portion and a judgment output from the level detecting means 83. This is a case where it is divided into switching means 82 and 84 which are switching portions for generating level signals. Therefore, since the level detecting means 83 is common among the channels, for example, after detecting the level of each channel, the low level signal is sent to all the channels at the same time in accordance with the low level signal period of the lowest level channel. As a result, the noise reduction efficiency can be increased as compared with the case where each channel independently performs a low level signal. Subsequent processing is the same as that in the example of FIG.

  Next, an example of FIG. 12 will be described. In the example of FIG. 1 described above, the noise reduction band becomes monaural. In the example of FIG. 12, noise reduction is performed while maintaining the sound field feeling of each input channel.

  First, the Rch signal and the Lch signal input from the input terminals 30 and 31 are respectively supplied to the level discriminating means 32 and 33, and are discriminated into the high level signal and the low level signal as described above. Is supplied to the adders 132 and 133, the Rch low level signal is supplied to the adder 34, the level value detection / determination means 126, and one fixed contact 128a of the changeover switch 128, and the Lch low level signal is added. , And the other fixed contact 129b of the level value detection / judgment means 127 and changeover switch 129.

  Further, the output of the adder 34 is multiplied by 1/2 by the 1/2 calculator 35 to obtain the level value detection / determination means 126 and 127 and the other fixed contact 128b of the changeover switch 128 and the changeover switch as an (L + R) ch signal. 129 is supplied to one fixed contact 129a.

  Here, the level value detection / determination means 126 and 127 have the same configuration as that of FIG. 4, but since the input is two channels, the smaller one of the level of the input signal is appropriately determined, and the changeover switches are respectively set. The Rch determination output and the Lch determination output are output to the 128 movable contacts 128c and 129 as the Rch determination output and the output determined by the changeover switches 128 and 129, respectively, and are selected to the adder 132 via the LPFs 130 and 131, respectively. The Rch signal is added to the high level signal of the lever level discriminating means 32 and outputted from the Rch output terminal 41. Similarly, the adder 133 adds it to the high level signal of the level discriminating means 33 and is sent from the Lch output terminal 42. An Lch signal is output.

  Further, the level discriminating means 32 and 33 may be implemented by separating the level detecting means and the switching means as shown in FIG.

  Here, the noise reduction apparatus in the example of FIG. 12 will be further described with reference to FIG. First, the low level signal from the level discriminating means 33 is the Lch signal in FIG. 16A, and the low level signal from the level discriminating means 32 is the Rch signal in FIG. 16B. Further, the adder 34 and the 1/2 calculator 35 generate the (L + R) / 2 composite signal of FIG. 16C. If the minimum value of the Lch signal in FIG. 16A and the (L + R) / 2 composite signal in FIG. 16C is always selected by the level value detection / judgment means 127 and the switching means 129, it is output as shown by the thick line in FIG. When the LPF 131 is passed through in order to suppress harmonic components, the output is as shown in FIG. 16E. Similarly, when the minimum value of the Rch signal in FIG. 16B and the (L + R) / 2 composite signal in FIG. 16C is always selected by the level value detection / judgment means 126 and switching means 128, it is output as shown by the thick line in FIG. Further, when the LPF 130 is passed through in order to suppress harmonic components, the output is as shown in FIG. 16G.

  The adders 132 and 133 add and re-synthesize the level, thereby generating an Rch signal and an Lch signal with reduced noise. Therefore, in the example of FIG. 12, the level of the random noise component can be reduced as shown in FIGS. 16E and 16G with respect to the (L + R) / 2 synthesized signal of FIG. 16C, which is an example of noise reduction of the example of FIG. In addition, the Lch signal component and the Rch signal component can be left without being made monaural.

  Here, a supplementary description will be given of the sampling interval for selecting the minimum value in this example and the subsequent band-limited frequency by the LPF. In FIG. 6D, FIG. 16D, and FIG. 16F described above, the time unit for selecting the minimum value by comparing the levels is the sampling interval that is the minimum time unit of the digital signal. At this time, the band limiting frequency in the subsequent stage may be set to Fs / 2 or less by the sampling theorem if the sampling frequency is Fs.

  So far, noise reduction in the two channels of Lch and Rch has been described, but it is also possible to deal with multichannels of three or more channels. The example of FIG. 13 shows three channel noise reduction. To explain the example of FIG. 13, Rch signal, C (Center) ch signal and Lch signal are inputted from input terminals 30, 141 and 31, respectively, and high level signals are respectively obtained by level discriminating means 32, 144 and 33. Each of the low level signals is divided into a low level signal, and each low level signal has a first fixed contact 37a, a second fixed contact 37e, a third fixed contact 37c, a level value detection / determination means 36a, and an adder 34a of the changeover switch 37s. Are added to each other by the adder 34a, averaged by the 1/3 calculator 35a, and the (L + R + C) / 3 composite signal as the (L + R + C) ch signal as the fourth fixed contact of the changeover switch 37s. 37f and level value detection / determination means 36a.

  Then, the level value detection / determination means 36a determines the signal having the lowest level every predetermined sampling period, and the signal is selected by the movable contact 37d of the changeover switch 37s, and each high level of each channel is passed through the LPF 38. The signal and the adders 39, 151 and 40 are combined in level and output from the output terminals 41, 155 and 42 as an Rch signal, a Cch signal and an Lch signal. Even in the case of four or more channels, the noise reduction processing can be performed by changing the averaging processing of each channel and similarly performing the minimum value selection processing.

  In the example of FIG. 13, as in the examples of FIG. 11 and FIG. 12, the level discriminating means is divided, and the average signal obtained by adding all channels and the minimum value selection process for each channel signal are performed. In addition, noise reduction processing that increases the independence of each channel may be performed.

  Next, the noise reduction device in the example of FIG. 14 will be described, but the description of the same function and the same block as the noise reduction device described above will be omitted. FIG. 14 shows an example in which the acoustic band is divided into band 1, band 2 and band 3 as shown in FIG. 17, and noise is reduced only to band 1 and band 2.

  First, the Rch signal and the Lch signal input from the input terminals 30 and 31 are band-divided by bandpass filters (BPF) 1 (163 and 164), 2 (165 and 166) and 3 (162 and 167), respectively. The noise reduction processing is performed for each of the divided bands of the band 1 and the band 2 by the BPF 1 (163 and 164 and BPF 2 (165 and 166). First, the output from the BPF 1 (163 and 164) of the Rch signal and the Lch signal is level discriminated. Supplied to means 32b and 33b.

  The low level signal from the level discriminating means 32b is supplied to the adder 34d, the level value detecting / determining means 36b and the first fixed contact 37a of the changeover switch 37s1, and the low level signal from the level discriminating means 33b is The adder 34d, the level value detection / determination means 36b and the third fixed contact 37c of the changeover switch 37s1 are supplied to the adder 34d and the (L + R) / 2 composite signal ((2) obtained from the 1/2 calculator 35b. The (L + R) ch signal is supplied to the level value detection / determination means 36b and the second fixed contact 37b of the changeover switch 37s1. The minimum value is selected by the level value detection / determination means 36b and the movable contact 37d of the changeover switch 37s1, and is supplied to the adder 180 via the LPF 178.

  Similarly, the outputs of the Rch signal and the Lch signal from the BPF 2 (165 and 166) are supplied to the level discriminating means 32c and 33c.

  The low level signal from the level discriminating means 32c is supplied to the adder 34e, the level value detecting / determining means 36c and the first fixed contact 37a of the changeover switch 37s2, and the low level signal from the level discriminating means 33c is The adder 34e is supplied to the third fixed contact 37c of the level value detection / determination means 36c and the changeover switch 37s2, and the (L + R) / 2 composite signal (L) is obtained from the adder 34e and the 1/2 calculator 35c. The (L + R) ch signal is supplied to the level value detection / determination means 36c and the second fixed contact 37b of the changeover switch 37s2. The minimum value is selected by the level value detection / determination means 36c and the movable contact 37d of the changeover switch 37s2, and is supplied to the adder 180 via the LPF 185.

  The adder 180 outputs a band-combined band 1 and band 2 noise reduction signal. Further, the output of the band 3 from the BPF 3 (162 and 167) of the Rch signal and the Lch signal and the high level signal from each level discriminating means 32b, 32c, 33b, 33c are respectively channeled by the adders 172, 173, 186, 187. The low level signals, which are synthesized every time and further subjected to noise reduction by the adders 188 and 189, are level synthesized, the Rch signal is output from the output terminal 41, and the Lch signal is output from the output terminal 42.

  By performing the minimum value selection processing for each band after performing band division in this way, it is possible to reduce noise that is a random component while improving the reproducibility of an audio signal that is the same phase component.

  In the example of FIG. 14 as well, as in the examples of FIG. 11 and FIG. 12 described above, the level discriminating means is divided, and the average signal obtained by adding all channels and the minimum value selection process for each channel signal are performed. In addition, noise reduction processing that increases the independence of each channel may be performed.

  Further, in the noise reduction apparatus in the example of FIG. 15, fast Fourier transform (FFT) is performed to further subdivide the band division than in the example of FIG. Here, the Rch signal and the Lch signal input from the input terminals 30 and 31 respectively convert the time axis signal in the acoustic band into n frequency axis signals of frequencies f1 to fn by the FFT means 304 and 306, respectively.

  Similarly, the (L + R) / 2 composite signal from the adder 34 and the ½ calculator 35 is also converted into n frequency axis signals of frequencies f1 to fn by the FFT means 305. Here, each FFT means separates the frequency axis signals of frequencies f1 to fn, and first, the Rch signal is separated by the FFT means 304 into a low level signal of a frequency signal smaller than a predetermined reference level (extraction level) and other high level signals. Then, the high level signal is supplied to the synthesizing unit 309, and the low level signal is supplied to the level comparing / selecting unit 307.

  Similarly, the Lch signal is also separated into a low level signal and other high level signals by the FFT means 306, the high level signal is supplied to the combining means 310, and the low level signal is supplied to the level comparison / selection means 308. The (L + R) / 2 composite signal is directly supplied from the FFT means 305 to the level comparison / selection means 307 and 308 as frequency axis signals of frequencies f1 to fn.

  Here, level comparison / selection for each frequency of frequencies f1 to fn will be described with reference to FIG. First, when the Rch signal is converted into the signals R1 to Rn having the frequencies f1 to fn as shown in FIG. 18A, for example, R2, R3, Rn below the predetermined reference level indicated by the broken line are L2, L3, Ln as shown in FIG. 18D. Are selected as low level signals, but at this time, the unselected R1 and Rn-1 signals are replaced with zero signals as L1 and Ln-1.

  18A, the R1 and Rn-1 signals are selected as high level signals as H1 and Hn-1 as shown in FIG. 18C. At this time, unselected R2, R3, and Rn are zero signals as H2, H3, and Hn. Is replaced by Here, when the signals A1 to An of the frequencies f1 to fn are output from the FFT means 305 as shown in FIG. 18B, the (L + R) / 2 composite signal is compared with the low level signal of FIG. The operation of selecting the signal with the lowest level as shown in FIG. 18E is performed for all the frequencies f1 to fn.

  That is, the signals L1, A2, L3, Ln-1, and An are selected. The signals selected in this way are input to the synthesizing means 309 and 310, and band synthesis is performed by level addition again with the high-side signals of the frequencies f1 to fn. An example of Rch output at this time is shown in FIG. 18F. H1, A2, L3, Hn-1, and An signals of frequencies f1 to fn are obtained. The frequency f1 to fn signals thus selected are sent to inverse fast Fourier transform (IFFT) means 311 and 312 and the frequency axis signal is inversely converted into a time axis signal to be output as Rch and Lch signals from the output terminals 42 and 42. Output.

  The example in FIG. 15 is an example, and can be changed within the scope of the object of the present invention described above.

  In general, in the case of a microphone input of a video camera, it is necessary to compress and record a natural sound in a predetermined dynamic range. Therefore, the input acoustic signal is level-compressed by an AGC (Automatic Gain Control) means. ing. That is, when the input acoustic signal is large, the gain is reduced by the AGC means, and when the input acoustic signal is small, the gain is increased by the AGC means. When the gain of this AGC means is small, that is, when the input is large, the level is suppressed and almost no sound can be heard. Conversely, when the gain is large, that is, when the surroundings are quiet, I can hear it clearly.

  Therefore, the AGC gain information from the AGC means is adaptively used for the level discriminating means in this example. For example, when the AGC gain is small, the high level signal is dominant by reducing the reference level to reduce the noise. If the AGC gain is large, if the AGC gain is large, the low level signal is dominant by increasing the reference level to increase the noise reduction effect. The effect can be optimized.

  Further, the series of noise reduction processes described above may be performed on a signal from a microphone as shown in FIG. 19 to configure a sound collection system or a recording system. However, the present invention is not limited to this, and reproduction is possible. It may be implemented in the system. Alternatively, it may be implemented as application software in a computer, and may be implemented as non-real time processing when editing a video / audio file, converting a file, or writing to a DVD disc.

  According to this example, the target noise is not correlated with each other in multiple channels and the phase is random noise, so it is reduced by averaging. In this example, in addition to this, the minimum selection of the input signal is selected. Since the phase difference component can be strongly removed by performing the processing, the reduction effect is increased even with a small number of microphones.

  In addition, according to this example, since only the noise level region below the predetermined level is processed, it can be reduced with little influence on the audio signal, and after the noise addition averaging process, the random noise is monaural signal in each channel. Therefore, for example, even if sound field calculation processing is performed from a plurality of channels in the subsequent stage, the noise level does not deteriorate.

  According to this example, noise can be reduced while leaving a sound field feeling (independence) by performing a minimum selection process on the noise addition average processing output and each acoustic channel output.

  According to this example, by dividing the band, performing the minimum selection process for each divided band, and performing band synthesis again in the final stage, it is possible to reduce the noise optimally for each band and to affect the acoustic signal. Less, and the reproducibility of the acoustic signal is improved.

  According to this example, it is effective in reducing white noise generated in the acoustic signal from the microphone, and is particularly effective in the pre-stage processing of the stereo sound field calculation processing and multi-channel (surround) sound field calculation processing.

  According to this example, it is also effective to reduce white noise generated in an audio signal during reproduction from a tape or disk medium in a video camera or the like.

  In this example, the minimum time unit for performing the minimum (minimum value) selection processing is an audio sampling time if it is a digital signal, and the extractable band in this case is the sampling frequency Fs by the sampling theorem (Nyquist theorem). If it does, it will be possible to Fs / 2. In general, if the high frequency of the noise band is up to Fn, the time unit for performing the minimum (minimum value) selection processing of this example is 1 / (2 · Fn) or more.

  In this example, the target noise is white noise that is at a relatively low level with respect to the acoustic signal and is present in almost the entire voice band. Therefore, by extracting a level region below a predetermined level, human noise is extracted. This makes it possible to reduce the noise using the masking effect, and to eliminate unnecessary noise reduction.

  In this example, by sharing the level detection means in each acoustic channel, each acoustic channel can control the switching means at the same time by avoiding noise extraction output individually according to the acoustic level. The noise reduction effect can be increased.

  According to this example, it is possible to reduce a sense of discomfort in the sense of hearing by performing the switching in the switching means relatively gradually with the cross-fade characteristic.

  According to this example, the low-level audio signal is slightly removed together with the white noise. By obtaining the level information of the input audio signal, the human auditory masking phenomenon is used, and only when the noise is noticeable. Noise reduction processing can be performed to minimize the influence on voice.

  Of course, the present invention is not limited to the above-described examples, and various other configurations can be adopted without departing from the gist of the present invention.

It is a block diagram which shows the example of the best form for implementing this invention noise reduction apparatus. It is a block diagram which shows the example of a level discrimination means. It is a diagram with which it uses for description of this invention. It is a block diagram which shows the example of a level value detection / determination means. It is a flowchart with which it uses for description of this invention. It is a diagram with which it uses for description of this invention. It is a block diagram which shows the other example of a level discrimination means. It is a block diagram which shows the other example of a level discrimination means. It is a block diagram which shows the example of a low level extraction circuit. It is a diagram with which it uses for description of FIG. It is a block diagram which shows the other example of the best form for implementing this invention noise reduction apparatus. It is a block diagram which shows the other example of the best form for implementing this invention noise reduction apparatus. It is a block diagram which shows the other example of the best form for implementing this invention noise reduction apparatus. It is a block diagram which shows the other example of the best form for implementing this invention noise reduction apparatus. It is a block diagram which shows the other example of the best form for implementing this invention noise reduction apparatus. It is a diagram with which it uses for description of the example of FIG. It is a diagram with which it uses for description of the example of FIG. FIG. 16 is a diagram for explaining the example of FIG. 15. It is a block diagram which shows the example of the conventional stereo sound field processing apparatus. FIG. 20 is a diagram for explaining FIG. 19.

Explanation of symbols

  30, 31 ... input terminals, 32, 33 ... level discrimination means, 34, 39, 40 ... adder, 35 ... 1/2 arithmetic unit, 36 ... level value detection / determination means, 37 ... changeover switch, 38 ... low-pass filter , 41, 42 ... output terminals

Claims (18)

Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extraction means for extracting a level region including a noise component in the acoustic signal; an arithmetic means for calculating an average of noise signals from the noise extraction means;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
A selection means for selecting a signal having the lowest level value among the level values detected from the first level detection means and the second level detection means for each predetermined period;
Band limiting means for limiting the band of the signal from the selection means;
A plurality of synthesizing units for synthesizing the signal from the band limiting unit and the signal other than the extraction level in the noise extracting unit for each acoustic channel;
The noise reduction apparatus characterized in that the output of each synthesizing means is each acoustic channel output signal. Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extracting means for extracting a level region including a noise component in the acoustic signal;
Arithmetic means for calculating an average of signals from the noise extraction means;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
A plurality of selection means for selecting a signal having a lower level value for each predetermined period in each acoustic channel from the level value from the first level detection means and the level value from the second level detection means. When,
A plurality of band limiting means for limiting the band of the signal from the selection means;
A plurality of synthesizing units for synthesizing the signal from the band limiting unit and the signal other than the extraction level in the noise extracting unit for each acoustic channel;
The noise reduction apparatus characterized in that the output of each synthesizing means is each acoustic channel output signal. Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
Band extraction means for extracting a predetermined band from the acoustic signal;
A plurality of noise extracting means for extracting a level region including a noise component in the signal from the band extracting means;
A plurality of computing means for computing an average of signals from the noise extracting means for each of the predetermined bands;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
Selection means for selecting a signal having a level value having the smallest level value from the first level detection means and the level value from the second level detection means for each predetermined period of the predetermined band;
Band limiting means for limiting the band of the signal from the selection means;
Band synthesis means for performing band synthesis for each acoustic channel and signals from the band limiting means and band signals other than the extraction band in the band extraction means;
A plurality of synthesizing units for synthesizing the signal from the band synthesizing unit and the signal other than the extraction level in the noise extracting unit for each acoustic channel;
The noise reduction apparatus characterized in that the output of the synthesizing means is each acoustic channel output signal. In the noise reduction device according to claim 1, 2, or 3,
The noise reduction apparatus, wherein the input means is a microphone. In the noise reduction device according to claim 1, 2, or 3,
The noise reduction apparatus according to claim 1, wherein the acoustic signal to the input means is a reproduction signal from a predetermined recording medium. In the noise reduction device according to claim 1, 2, or 3,
The noise reduction apparatus according to claim 1, wherein the minimum unit constituting the predetermined period is a sampling period. In the noise reduction device according to claim 1, 2, or 3,
The noise reduction apparatus, wherein the noise component is white noise ((phase) random noise). In the noise reduction device according to claim 1, 2, or 3,
The noise extraction means is constituted by a level detection means and a switching means, and when the detection level in the level detection means is smaller than a predetermined level set separately, the noise extraction means outputs the noise. A noise reduction device. The noise reduction device according to claim 8,
When the detection level of the acoustic signal from the plurality of acoustic channels is smaller than the predetermined level in any one channel, or the sum of the detection levels of the acoustic signals is smaller than the predetermined level In addition, a noise reduction apparatus characterized in that the switching means performs noise extraction output on all acoustic channels. The noise reduction device according to claim 8,
The noise reduction device characterized in that the switching means has a crossfade switching characteristic. The noise reduction device according to claim 8,
The noise reduction apparatus according to claim 1, wherein the predetermined level is based on level information from AGC means. The noise reduction device according to claim 3,
The noise reduction apparatus according to claim 1, wherein the band extracting unit includes a plurality of filter units or FFT units. An input step of inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extraction steps for extracting a level region including a noise component in the acoustic signal;
A calculation step of calculating an average of noise signals from the noise extraction step;
A plurality of first level detection steps for detecting a signal level in a predetermined period of the signal from the noise extraction step;
A second level detection step of detecting a signal level in a predetermined period of the signal from the calculation step;
A selection step of selecting a signal having the lowest level value among the level values detected from the first level detection step and the second level detection step for each predetermined period;
A band limiting step for limiting the band of the signal from the selection step;
A plurality of synthesis steps for synthesizing the signal from the band limiting step and the signal other than the extraction level in the noise extraction step for each acoustic channel;
The noise reduction method characterized in that the output of each synthesis step is used as each acoustic channel output signal. An input step of inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extraction steps for extracting a level region including a noise component in the acoustic signal;
A calculation step of calculating an average of signals from the noise extraction step;
A plurality of first level detection steps for detecting a signal level in a predetermined period of the signal from the noise extraction step;
A second level detection step of detecting a signal level in a predetermined period of the signal from the calculation step;
A plurality of selection steps for selecting a signal having a lower level value for each predetermined period in each acoustic channel from the level value from the first level detection step and the level value from the second level detection step. When,
A plurality of band limiting steps for limiting the band of the signal from the selection step;
A plurality of synthesis steps for synthesizing the signal from the band limiting step and the signal other than the extraction level in the noise extraction step for each acoustic channel;
The noise reduction method characterized in that the output of each synthesis step is used as each acoustic channel output signal. An input step of inputting a plurality of acoustic signals from a plurality of acoustic channels;
A band extracting step of extracting a predetermined band from the acoustic signal;
A plurality of noise extraction steps for extracting a level region including a noise component in the signal from the band extraction step;
A plurality of calculation steps for calculating an average of signals from the noise extraction step for each of the predetermined bands;
A plurality of first level detection steps for detecting a signal level in a predetermined period of the signal from the noise extraction step;
A second level detection step of detecting a signal level in a predetermined period of the signal from the calculation step;
A selection step of selecting a signal having a level value having the smallest level value from the first level detection step and the level value from the second level detection step for each predetermined period of the predetermined band;
A band limiting step for limiting the band of the signal from the selection step;
A band synthesis step of performing band synthesis for each acoustic channel with a signal from the band limiting step and a band signal other than the extraction band in the band extraction step;
A plurality of synthesis steps for synthesizing the signal from the band synthesis step and the signal other than the extraction level in the noise extraction step for each acoustic channel;
The noise reduction method characterized in that the output of the synthesis step is each acoustic channel output signal. Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extraction means for extracting a level region including a noise component in the acoustic signal; an arithmetic means for calculating an average of noise signals from the noise extraction means;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
A selection means for selecting a signal having the lowest level value among the level values detected from the first level detection means and the second level detection means for each predetermined period;
Band limiting means for limiting the band of the signal from the selection means;
A plurality of synthesizing means for synthesizing the signal from the band limiting means and the signal other than the extraction level in the noise extracting means for each acoustic channel;
A noise reduction program characterized in that an output of each synthesis means is used as an output signal of each acoustic channel. Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
A plurality of noise extracting means for extracting a level region including a noise component in the acoustic signal;
Arithmetic means for calculating an average of signals from the noise extraction means;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
A plurality of selection means for selecting a signal having a lower level value for each predetermined period in each acoustic channel from the level value from the first level detection means and the level value from the second level detection means. When,
A plurality of band limiting means for limiting the band of the signal from the selection means;
A plurality of synthesizing means for synthesizing the signal from the band limiting means and the signal other than the extraction level in the noise extracting means for each acoustic channel;
A noise reduction program characterized in that an output of each synthesis means is used as an output signal of each acoustic channel. Input means for inputting a plurality of acoustic signals from a plurality of acoustic channels;
Band extraction means for extracting a predetermined band from the acoustic signal;
A plurality of noise extracting means for extracting a level region including a noise component in the signal from the band extracting means;
A plurality of computing means for computing an average of signals from the noise extracting means for each of the predetermined bands;
A plurality of first level detection means for detecting a signal level in a predetermined period of the signal from the noise extraction means;
Second level detection means for detecting a signal level in a predetermined period of the signal from the calculation means;
Selection means for selecting a signal having a level value having the smallest level value from the first level detection means and the level value from the second level detection means for each predetermined period of the predetermined band;
Band limiting means for limiting the band of the signal from the selection means;
Band synthesis means for performing band synthesis for each acoustic channel and signals from the band limiting means and band signals other than the extraction band in the band extraction means;
A plurality of synthesizing units for synthesizing the signal from the band synthesizing unit and the signal other than the extraction level in the noise extracting unit for each acoustic channel;
A noise reduction apparatus program characterized in that the output of the synthesizing means is each acoustic channel output signal.

Images