JP3823594B2 - Mechanical noise reduction device and mechanical noise reduction method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a noise reduction apparatus and a noise reduction method for reducing mechanical noise (noise generated from a mechanical part and mixed in an audio signal).
[0002]
[Prior art]
Mechanical noise reduction system that reduces only the head tapping sound when noise sound caused by contact between the rotating head of a camera-integrated VTR such as a digital camcorder (trade name) and the tape, so-called head tapping sound, is mixed into the audio signal from the microphone. Has already been proposed.
[0003]
A conventional mechanical noise reduction method will be described below with reference to FIG.
The conventional mechanical noise reduction method shown here is a device for reducing continuously generated noise, and the noise source is a rotating drum.
[0004]
In the figure, microphones MIC1A and MIC1B are stereo microphones built in the apparatus main body, but this need not be a stereo microphone, and may be a monaural microphone. The output signals 1L and 1R of MIC1A and MIC1B are amplified by the AMP2A and AMP2B preamplifiers to become signals 2L and 2R, adjusted to optimum levels by the AGC (automatic gain control) circuit 3 to become signals 3L and 3R, and AD converters ( (Hereinafter referred to as ADC4A and ADC4B) are converted into digital signals 4L and 4R.
[0005]
The digital signals 4L and 4R are input to the mechanical noise cancellation processing circuit 15, and the adder 6A and 6B subtract the pseudo noise signals 5L and 5R for each of the left L and right R channels and output as signals 6L and 6R. . The signals 6L and 6R are processed as error signals that are fed back to the adaptive filters 5A and 5B. The adaptive filters 5A and 5B use the error signal and a drum reference signal 7 input from a microcomputer (hereinafter referred to as a microcomputer) 8 to perform an audio sampling frequency (32 kHz, 44.1 kHz) by the LMS method (least square method). (Or 48 kHz) so that the error signal is minimized, and signals 5L and 5R are output.
[0006]
The drum reference signal 7 is a reference for the rotation phase and frequency of the drum motor 13 that is servo-controlled by the servo signal 12 sent from the same microcomputer 8, so that the rotation when driven by the drum motor 13 is performed. The drum noise sound caused by the drum 14 and the head noise noise caused by the contact between the rotating magnetic heads 11A and 11B attached to the rotating drum 14 and the magnetic tape 9 are also generated in the harmonic region of the drum reference signal 7. Therefore, this noise is transmitted by a transfer function through the space or the cabinet of the equipment and is input to the built-in microphone together with the sound. The output of the mechanical noise cancellation processing circuit 15 is sent to the recording system signal processing and recorded on the magnetic tape 9 together with the video signal. However, the description of this part is omitted because it is not directly related to the present invention.
[0007]
[Problems to be solved by the invention]
This mechanical noise reduction method is a very effective noise reduction method when constant noise is generated continuously in time. However, when the noise changes due to mode transition or the like, or intermittent in time. This is not a very effective noise reduction method in the case of a noise sound that is generated. In the latter case, multiple types of noise sounds are generated continuously, or noise generation is short-lived, and there is not enough time to generate and correct pseudo-noise. As a result, noise cannot be reduced. There is a case.
[0008]
An example of noise generation in the latter case will be described below.
1. Noise caused by disk rotation sound, head movement sound for search and writing, and stepping motor sound, etc. when recording a sound simultaneously with an image with a still image camera using a floppy disk, MD or the like as a recording medium .
2. Noise caused by motor sound during autofocus of a camera-integrated VTR.
3. Recording REC mode and recording / pause REC of camera-integrated VTR Noise generated at the time of mode transition such as PAUSE mode repetition.
[0009]
In the conventional noise reduction method using an adaptive filter, when noise actually occurs, the noise is compared with the pseudo noise generated by the adaptive filter, and the error component that is the difference is input to the adaptive filter again. Thus, since a pseudo-noise is generated so that an error component is eliminated by configuring a feedback loop, there is a problem that it takes time to pull up to a target value when an error is large.
[0010]
The present invention aims to allow efficient noise reduction to noise intermittently occurs or when a plurality of kinds of noise.
[0011]
[Means for Solving the Problems]
The noise reduction device of the present invention is a noise reduction device for obtaining an audio signal in which only a noise component is reduced from an audio signal mixed with noise generated in a predetermined period which is not continuous in time but intermittent. A noise signal similar to the noise generated in this predetermined period, which is picked up from the microphone, is recorded in the memory as a pseudo noise waveform. The noise is read out in synchronization with the noise pitch signal including the noise and subtracted from the input signal to reduce the noise.
[0012]
The noise reduction apparatus of the present invention is the noise reduction apparatus according to claim 1, wherein a plurality of types of pseudo noise are stored in the memory, and the capacity of the memory is included in one cycle of the repetition period of the noise pitch signal. If the number of addresses is the same as the number of addresses, and noise reduction is performed from the input signal, when reducing multiple types of noise components, the memory capacity corresponding to the maximum pitch of the noise components should be secured. In addition, by making the address control of this memory variable according to the noise pitch signal, a plurality of types of noise noise can be reduced.
[0013]
The noise reduction device according to the present invention has the learning effect of correcting the difference between the noise waveform from the current microphone and the pseudo noise waveform by the LMS method (least square method) during the noise reduction operation. It is what you have.
[0014]
The noise reduction method of the present invention is a noise reduction method for obtaining an audio signal in which only a noise component is reduced from an audio signal mixed with noise generated in a predetermined period which is not continuous in time but intermittent. A noise pitch signal including a start ID having the same phase as the noise is obtained by sampling a noise signal in advance and recording it in a memory as a pseudo noise waveform, and collecting the pseudo noise similar to the noise generated during the predetermined period collected from the microphone. The noise is read out from the memory in synchronism with the signal and subtracted from the input signal.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
Next, referring to FIG. 1, a description will be given of an embodiment of a noise reducing apparatus of the present invention.
Here, the description of the same parts as those in FIG. 3 is omitted.
The signals 4L and 4R from the ADC 4 are input to a mechanical noise cancellation processing circuit 33 constituted by a DSP (digital signal processor), ASIC_IC, or the like.
[0016]
The signals 4L and 4R are subtracted from the pseudo noise signals 20L and 20R output from the adaptive filters 20A and 20B independently from the L and R channels by the adders 23A and 23B, and are output as 23L and 23R. To be recorded on various recording media. In the example shown in FIG. 1, the signals 23L and 23R are recorded on the disk 27. The subsequent circuit for processing the signals 23L and 23R is not directly related to the present invention, and thus detailed description thereof is omitted. The signals 23L and 23R are input as error signals to the adaptive filters 20A and 20B independent of the left L and right R channels, respectively. The adaptive filters 20A and 20B are basically configured by LMS processing. Here, a signal from the volatile memory SRAM (static RAM) 21A and 21B and a pitch signal 24 sent from the microcomputer 25 are included. input.
[0017]
Here, the SRAMs 21A and 21B are connected to the adaptive filters 20A and 20B through a bidirectional communication path so that data can be read and written and address control can be performed by the read / line signal 22. Further, exchange of signals 26L and 26R with the nonvolatile memory EEPROM 26 built in the microcomputer 25 is performed. Signals 26L and 26R indicate bidirectional communication signals, and data can be exchanged between the EEPROM 26 and the SRAMs 21A and 21B. The pitch signal 24 is a noise reference pulse and includes, for example, a minimum frequency pitch in the same phase as the generated noise and a start ID. Detailed description of the adaptive filters 20A and 20B will be given later. In FIG. 1, an EEPROM is used as the nonvolatile memory, but a nonvolatile memory such as a flash memory may be used.
[0018]
The microcomputer 25 sends a control signal to the disk motor 28 and the step motor 30 which are noise generation sources. That is, a motor control signal 32 is sent to the disk motor 28, and a motor control signal 31 is sent to the step motor 30 for head position control. Each of these control (control) signals performs control for rotating the motor at a constant speed and position control for moving the head to a target track or sector. Since the disk driven by the disk motor 28 and the disk drive system operate only when accessed from the microcomputer 25, noise noise is generated intermittently.
[0019]
The same applies to the head driven by the step motor 30, and noise noise is generated intermittently. In the example of FIG. 1, the disk is taken as an example, but the present invention can be similarly applied to an AF (auto focus) control of a video camera controlled by a microcomputer.
[0020]
Here, the noise reduction apparatus of the present embodiment samples the generated noise sound in advance, but in general, the noise sound is a multitone signal that includes signals of a plurality of frequencies and levels in a complicated manner. Therefore, it is difficult to generate artificially.
[0021]
Therefore, in the noise reduction apparatus according to the present embodiment, as an example, pseudo noise is generated by sampling an actual noise sound according to the following procedure.
1. A device (product) is prepared in which the structural position of the noise source and the microphone and the mechanical structure that affects sound transmission such as the cabinet are determined.
2. Prepare an environment where only noise is input to the microphone. For example, the device is operated in an acoustic anechoic room (a room where there is no external sound or reflected sound) and only noise noise is generated.
3. Here, since only the noise sound is output as the signals 23L and 23R to the outputs of the adders 23A and 23B in FIG. 1, this is input to the adaptive filters 20A and 20B, and at least one pitch of noise to be generated is generated. The noise waveform of the minute is taken into the memory SRAMs 21A and 21B in synchronization with the pitch signal 24.
4). Since the SRAMs 21A and 21B are volatile memories, they are sent to the microcomputer 25 using the signals 26L and 26R and stored in the nonvolatile memory EEPROM 26 in the microcomputer.
[0022]
In this way, by capturing the noise waveform for at least one pitch in the memory, it is only necessary to repeatedly read from the memory for the time of the generated noise in synchronization with the pitch signal 24 thereafter. One pitch here corresponds to, for example, one rotation of a motor that is a noise generation source.
[0023]
Next, the adaptive filters 20A and 20B and the SRAMs 21A and 21B in FIG. 1 will be described in more detail below with reference to FIG.
[0024]
In the circuit of FIG. 2, a signal 41 input from the upper left of the figure is a signal in which an audio signal and a noise signal are mixed, and is applied to one input of the adder 45. The adder 45 subtracts the pseudo noise signal 60 sent from the adaptive filter 20 applied to the other input from the signal 41 and outputs it as a signal 46 consisting only of the audio signal.
[0025]
In the subtraction in the adder 45, as a result of subtracting the pseudo noise signal 60 from the signal 41, the noise signal included in the signal 41 may not be completely removed and may remain in the output signal 46. In the circuit of FIG. 2, a path for feeding back a part of the output signal 46 to the adaptive filter is provided. Therefore, when a noise signal component remains in the output signal 46, a signal including the noise signal component is included. Is taken into the adaptive filter 20, and the limiter 47 provided in the filter 20 limits the large level portion of the audio signal, and then passes through the gain control circuit 48 and is sent to one input of the adder 49. . This signal is added to the data signal 54 read from the SRAM 21 applied to the other input of the adder 49 and written to the SRAM 21 as the data output 53. In the above operation, the audio signal included in the signal 46 is also input to the adaptive filter 20, but in most cases, it is limited by the limiter 47.
[0026]
The data signal 54 read from the SRAM 21 passes through the switch SW51 and is sent to the other input of the adder 45 as the pseudo noise signal 60. As a result, a noise reduction operation loop is formed. Here, the gain control circuit 48 can change the gain by the microcomputer control signal 44. The switch SW51 can control ON / OFF of the pseudo noise signal 60 sent to the adder 45 by a cancel control signal 42 sent from a microcomputer or the like.
[0027]
The address address for reading and writing to the SRAM 21 is designated by the address control signal 52 sent from the adaptive filter 20. Thus, the address control signal 52 is generated by the read / line address counter 50 in the adaptive filter 20. The read / line address counter 50 is subjected to address reset by the pitch signal 43, and read / write is repeatedly executed.
[0028]
The memory capacity of the SRAM 21 is a memory having a word length (W (n) in FIG. 2) corresponding to the number of addresses included in one cycle of the pitch signal 43. For example, when the reduction of the sub-type noise components is switched and executed, it is necessary to secure a memory capacity corresponding to the maximum pitch of the noise components.
[0029]
The SRAM 21 is also provided with an input / output port from the microcomputer, and is provided with an address control signal 55, an input port for the data input 56, and an output port for the data output 57. With respect to the input / output port from the adaptive filter 20 at the normal time, this microcomputer input / output port is input / output, for example, at the time of power ON / OFF of the device or mode transition.
[0030]
When sampling the input noise sound to create the pseudo noise, the gain of the gain control circuit 48 is set to 1 (through state), the switch SW51 is turned off (off), and input to the adaptive filter 20. This is realized by directly writing the noise sound to the SRAM 21. In addition, when noise is intermittently generated, the switch SW51 is turned on when the motor that is a noise generation source is rotating and turned off when the motor is stopped, so that noise is reduced only when noise is generated. I have to.
[0031]
In the present invention, the following application is possible by adopting the circuit configuration as described above.
1. By storing multiple types of presumed noise sounds in the EEPROM of the microcomputer, it is possible to reduce noise even when the noise sounds change during the mode transition or because of the use of multiple motors. is there.
2. By making the address control of the SRAM 21 variable according to the pitch signal 43, noise noise depending on a plurality of types of motor rotation speeds (rotation pitches) can be reduced.
[0032]
Next, a second embodiment of the present invention will be described.
Since the adaptive filters 20A and 20B in FIG. 1 operate with the LMS algorithm at the time of normal noise reduction, the contents of the SRAM 21 are always updated to minimize the difference between the actual noise sound and the subtracted pseudo noise sound. . The update cycle is performed for each sampling of audio sampling frequencies of 32 kHz, 44.1 kHz, and 48 kHz, for example.
[0033]
Here, in the LMS processing, the data W in the SRAM 21 is updated by the following arithmetic expression.
W (n) = W (n−1) +2 μ · E (n−1) · X (n−1)
However, n is the number of samples, μ is an update step gain, which corresponds to the gain control circuit 48 in FIG. 2, E is a residual error signal, which corresponds to the audio signal output 46 in FIG. 2, and X is a reference input. This corresponds to a pitch signal 43 of 2.
[0034]
Therefore, since the noise reduction apparatus according to the present embodiment constantly updates the data in the SRAM 21 to the latest noise waveform and stores it in the memory while the power is on, for example, even when intermittent noise occurs. Subtraction can be started from the previous latest noise waveform, and reduction efficiency is good. When the power is turned off, the noise waveform immediately before turning off is written in the EEPROM 26 in the microcomputer to update the data. When the power is turned on next time, the data is read from the EEPROM 26 to the SRAM 21 again to reproduce the noise waveform. Subtraction can be started from the noise waveform.
As described above, in the present invention, it is possible to give a learning effect to noise reduction. For example, noise can be reduced even by a change in noise due to aging of the device.
[0035]
【The invention's effect】
In the conventional noise reduction method using an adaptive filter, when noise actually occurs, the noise is compared with the pseudo noise generated by the adaptive filter, and the error component that is the difference is input to the adaptive filter again. Thus, since a pseudo-noise is generated so that an error component is eliminated by configuring a feedback loop, there is a problem that it takes time to pull up to a target value when an error is large.
In contrast, noise reduction apparatus of the present invention, the noise target value noise is sampled in the memory at the time that occurred as a pseudo-noise in the initial state, so as to subtract this from the input signal Therefore, there is an effect that almost no time painter draws in the drawing by the adaptive filter.
[0036]
Moreover, noise reduction apparatus of the present invention, as described above, with a pre-mechanical noise signal sampling, advance allowed stored in a nonvolatile memory such as a pseudo noise waveform memory in noise pitch mechanical noise which is picked up by the microphone Since the noise is reduced by reading out the pseudo noise from the input signal and subtracting it from the input signal, even the intermittent noise can be reduced quickly, and the noise can be reduced in response to noise changes such as mode transitions. .
[0037]
Furthermore, by providing a learning effect that corrects the difference between the current noise waveform and the pseudo noise waveform by the LMS method (least square method) during the reduction operation, the reduction effect at the next noise occurrence can be further improved. In addition, it can cope with noise waveform changes due to aging of equipment.
The noise reduction device of the present invention is advantageous for future miniaturization of equipment because the built-in microphone and various noise sources can be arranged close to each other.
In addition, since the noise reduction device of the present invention can be calculated by digital processing, it is advantageous in future system LSI implementation.
[Brief description of the drawings]
FIG. 1 is a circuit block diagram showing an example of a mechanical noise reduction apparatus of the present invention.
FIG. 2 is a circuit block diagram of an adaptive filter and SRAM portion of the apparatus of the present invention.
FIG. 3 is a circuit block diagram of a conventional mechanical noise reduction device.
[Explanation of symbols]
MIC1A, MIC1B ... Microphone, AMP2A, AMP2B ... Amplifier, 3 ... AGC circuit, ADC ... Analog to digital converter, 20A, 20B ... Adaptive filter, SRAM ... Memory, 25 ... Microcomputer, 26 ... ... EEPROM (memory), 28 ... disk motor

Claims (4)

A noise reduction device for obtaining an audio signal in which only a noise component is reduced from an audio signal mixed with noise generated in a predetermined period that is not continuous in time,
It has means for sampling a presumed noise signal in advance and recording it in a memory as a pseudo noise waveform,
By reading from the memory and subtracting from the input signal a pseudo noise similar to the noise generated during the predetermined period collected from the microphone in synchronization with the noise pitch signal including the start ID in the same phase as the noise. A noise reduction device that performs noise reduction. The noise reduction device according to claim 1,
A plurality of types of pseudo noise are stored in the memory, and the capacity of the memory is a word length corresponding to the number of addresses included in one cycle of the repetition cycle of the noise pitch signal, and noise reduction from the input signal is performed. When switching between multiple types of noise component reduction, the memory capacity corresponding to the maximum pitch of the noise component is secured, and the address control of the memory can be varied according to the noise pitch signal. By doing so, a noise reduction device characterized in that a plurality of types of noise noise can be reduced. The apparatus of claim 1.
A noise reduction apparatus characterized by having a learning effect of correcting a difference between a noise waveform from a current microphone and a pseudo noise waveform by an LMS method (least square method) during the noise reduction operation. A noise reduction method for obtaining an audio signal in which only noise components are reduced from an audio signal mixed with noise generated in a predetermined period that is not continuous in time,
Sample the expected noise signal in advance and record it in memory as a pseudo noise waveform
By reading from the memory and subtracting from the input signal a pseudo noise similar to the noise generated during the predetermined period collected from the microphone in synchronization with the noise pitch signal including the start ID in the same phase as the noise. A noise reduction method comprising performing noise reduction.

Images