JP2708913B2 - Sound detection output device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention detects the presence or absence of a valid signal component such as voice included in an input signal, and inputs this to a voice recognition device, a voice storage device, and the like. The present invention relates to a voice detection output device.

[Conventional technology]

Conventionally, a voice detection device for capturing an input voice in voice recognition, voice storage, and the like is described in Japanese Patent Application Laid-Open No. 63-291096, for example, when voice is input to a voice recognition device using a microphone or the like. As has been
The input voice section is detected using the power and the threshold value calculated from the microphone output signal, and the detected voice is input to the voice recognition device.

 Hereinafter, this will be briefly described with reference to FIG.

FIG. 7 is an explanatory diagram of the principle of detecting a speech section from an input signal, and is a waveform diagram of an output signal of a microphone in which the horizontal axis represents time and the vertical axis represents power.

In the drawing, when the power of the microphone output signal becomes larger than a threshold value TS, it is determined that a voice is detected. Although using this power and threshold to detect speech is very simple, it has many disadvantages. For example, if there is a low power consonant at the beginning,
If the threshold value TS is set large, a consonant at the beginning of a word is dropped to detect a voice section. In order to detect a speech section without dropping the consonant at the beginning of the word, the threshold value TS may be reduced. There is.

In order to prevent this, in the invention of Japanese Patent Application Laid-Open No. 63-291096, voice detection is determined based on the above-mentioned power and threshold value, and a change in the inter-vector distance of a characteristic parameter of the input signal is used as auxiliary information and the threshold value. The voice detection is performed in consideration of the judgment of the above. this is,
Characteristic parameters of noise and voice are largely different, and when a voice is present in noise, the distance between vectors of the characteristic parameter of the input signal changes.

However, extracting feature parameters requires large hardware.

Further, as an improvement measure of voice detection based on power and a threshold, as in the invention described in JP-A-64-80999, an input signal is divided into bands and each power is determined by a threshold, or The threshold value can be changed, and another microphone dedicated to collecting ambient noise
There is a method in which the noise level is measured, and the threshold value is set based on the measured noise level.

[Problems to be solved by the invention]

The above-mentioned prior art is based on performing voice detection based on the power of an input signal and a threshold, but does not mention how to set the threshold. Also, measure the noise level only with another dedicated microphone,
With this setting of the threshold value, the other dedicated microphone cannot be spatially shielded from voice input, so that voice may be input and malfunction may occur. further,
No consideration is given as to when the noise level measurement will be performed.

In the case of voice input, since the surrounding environment, that is, the surrounding noise is changing every moment, accurate detection of a voice section is impossible unless the threshold level is constantly changed.

As described above, in the related art, no consideration is given to the setting of this threshold value, that is, what value should be set at what time, and the input to the speech recognition device is not accurate. However, there were problems such as a decrease in the recognition rate.

An object of the present invention is to set the threshold value to an optimum value at an optimum timing, thereby accurately detecting a voice section, and inputting a voice to a voice recognition device or a voice storage device to perform recognition. It is an object of the present invention to provide a voice detection output device capable of preventing a reduction in rate or malfunction and reducing a file cost in storage.

[Means for solving the problem]

The above object is to provide a reminder sound output means for outputting a sound for informing a voice input to a speech recognition or storage device as a sound wave to a space, a microphone installed in the space to collect the sound, and a signal output by the microphone. A sound signal erasing means for erasing a sound signal output from the sound signal output means, a power detection means for detecting the power of a microphone output signal, and a power signal output means for detecting the power of the microphone output signal. A threshold generating means for generating and holding a threshold value, a comparing means for comparing outputs of the power detecting means and the threshold generating means, and detecting a voice section after the signal of the reminder sound; This is achieved by forming a voice detection output device with opening / closing means for controlling the output of a microphone output signal by the output of the microphone and connecting to a voice recognition or storage device.

[Action]

The notification sound output means emits an electronic sound or a voice by voice synthesis from a speaker in order to give a voice input to the operator. This sound is input to the microphone and becomes a microphone output signal. At this time, ambient noise is also input to the microphone at the same time. The reminder sound erasing means deletes the reminder sound component from the microphone output signal. The power detection means detects the power of the microphone output signal while the audible sound is being output. This power is detected as the power of the ambient noise by the function of the warning sound canceling means. The threshold value generating means receives the output of the power detecting means at this time, that is, the power of the ambient noise, generates a threshold value proportional thereto, and holds the threshold value. When the audible signal ends, the operator speaks for voice recognition or inputting voice to the storage device. The microphone output signal at this time is a mixture of the voice of the operator and the ambient noise. The comparing means compares the power output from the power detecting means at this time with the threshold value of the threshold value generating means set and held at the time of the preceding sound, and the power becomes larger than the threshold value. At times, the opening / closing means is closed to output a microphone output signal to the voice recognition or voice storage device.

The threshold value at this time is proportional to the ambient noise power immediately before the occurrence of the operator.Considering that the utterance of the operator is always higher than the ambient noise level as a person, the comparison means is Accurately comparing and determining the ambient noise and the voice, the voice is detected, and the voice signal of the operator collected by the microphone is output to the voice recognition or voice storage device.

In other words, in an environment that changes every moment, the detection level of the sound can be set as the optimum value immediately before the sound signal, which is the optimum timing.

By the above operation, the voice recognition or the storage device can accurately input the operator's voice regardless of the ambient noise level, so that the recognition rate can be reduced or malfunctions can be prevented, and the file cost for voice storage can be reduced. .

〔Example〕

 Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram for explaining a first embodiment of a voice detection and output device according to the present invention, wherein a is a voice detection and output device, and b is a voice recognition device.

In FIG. 1, reference numeral 1 denotes an alarm sound output circuit for outputting a sound informing the operator of a voice input to the apparatus, 2 a microphone for collecting the voice of the operator, and 3 a band elimination filter (hereinafter referred to as a band elimination filter) for removing a specific frequency component. , BEF), 4
Is a selection switch for selecting one of the output signal of the microphone 2 and the signal that has passed through the BEF 3, 5 is a power detection circuit for detecting the power of the signal, and 6 is a threshold value that is proportional to the output of the power detection circuit. A threshold generation circuit for generating and holding as a reference, 7 a comparison circuit for comparing the outputs of the power detection circuit 5 and the threshold generation circuit 6, 8 a switch for opening and closing the microphone output signal, and 9 a control of the entire apparatus. The control circuit 10 performs an interface with an external device. The voice detection output device a is
~ 10 circuits.

The alarm sound output circuit 1 includes a sine wave generation circuit 101 and an amplification circuit 1
02 and a speaker 103. Sine wave generation circuit 101
Generates a sine wave having an audible frequency f ₁ , amplifies the sine wave with the amplifier circuit 102, and outputs the sine wave from the speaker 103.

BEF3 intended to remove only the signal component of the frequency f _1, the other frequency band components to pass as it is. The power detection circuit 5 includes a rectifier circuit, a smoothing circuit, and the like, and outputs an average power of a signal. Of course, a more complex circuit configuration that outputs the effective power of the signal may be used.

FIG. 2 is a timing chart for explaining the operation of the embodiment shown in FIG. 1, and the operation of this embodiment will be described accordingly.

The degree of achievement of the current speech recognition technology is low, and many practical speech recognition devices recognize words uttered by a specific speaker. And few of them are designed to accept voice at any time, and most of them give voice to the operator by some kind of signal, for example, guidance by electronic sound or voice synthesis, and the voice generated within a certain period of time thereafter is transmitted to the operator. And recognizes it. Nevertheless, if the ambient environment, especially the ambient noise level, changes, it is a reality that the voice capture fails and the recognition rate decreases. The present invention has been made to prevent such a malfunction.

The voice recognition device b first issues a command sound generation command to the control circuit 9 via the interface circuit 10 to the voice detection output device a. The control circuit 9 controls the alarm sound output circuit 1, generates a sine wave having an audible frequency f ₁ for a predetermined time in the sine wave generation circuit 101, amplifies the sine wave in the amplification circuit 102, and outputs the sound from the speaker 103.

At this time, the control circuit 9 simultaneously controls the selection switch 4 for a certain period of time, and guides the signal collected by the microphone 2 and passed through the BEF 3 to the power detection circuit 5. Outside of this time, the signal collected by the microphone 2 is input to the power detection circuit 5 as it is.

At this time, the sound of the audible frequency f ₁ and the ambient noise are input to the microphone 2, but the BEF 3 is collected by the microphone 2 by the power detection circuit 5 to remove the signal component of the audible frequency f _1. Only the signal due to the ambient noise among the noises is input.

One point c in a fixed time during which this sound is output
Then, the control circuit 9 controls the threshold value generation circuit 6 to reset a threshold value proportional to the power value output from the power detection circuit 5 and hold it thereafter (the threshold value generation circuit 6
Until this time point c, the threshold value set in the same manner as in the case of the preceding sound output is held.)

In the above description, the time length of the sound output is set by the control circuit 9. However, the present invention is not limited to this. The voice recognition device b specifies the time length via the interface circuit 10, and the control circuit 9 specifies the time according to the specification. You may comprise so that the time length of a sound output may be set.

The time point c for resetting the threshold value by controlling the threshold value generating circuit 6 may be any time during which the audible alarm is being generated, but is preferably as close to the end of the audible alarm as possible.

The operator emits a sound after hearing the audible sound. At this time, the ambient noise continues to be input to the microphone 2.

The comparison circuit 7 includes a power detection circuit 5 and a threshold generation circuit 6
Is compared with the newly set threshold value, the time when the output signal of the power detection circuit 5 becomes larger than this threshold value is detected as the operator utterance voice section, and the switch 8 is closed. Then, the voice uttered by the operator is output to the voice recognition device b.

As described above, in this embodiment, the ambient noise level is measured during the output of the audible sound, that is, immediately before the utterance of the operator, whereby one threshold of the comparison circuit 7 for voice detection is set. Since the operator's utterance is detected based on the threshold value, the voice section can be reliably detected even when the ambient noise level fluctuates. In addition, it is possible to prevent the recognition rate of the voice recognition device from lowering and malfunction.

Further, in this embodiment, the audible sound is a single sine wave, but the audible sound is not limited to this and may be a mixture of sine waves of multiple frequencies. For example, if it is a DTMF tone, the operator does not feel uncomfortable. Of course, in this case, BEF3 must remove this multi-frequency component.

Although the comparison circuit 7 is always operating in the above description, it is preferable that the control circuit 9 controls the comparison circuit 7 to perform the comparison operation only for a certain period of time after the end of the audible signal. Further, it is preferable that this fixed time is the same as the voice input reception time of the voice recognition device 6.

FIG. 3 is a block diagram for explaining a second embodiment of the present invention, and the same reference numerals as those in FIG. 1 denote the same components.

In the figure, reference numeral 11 denotes an averaging circuit which averages the power value output from the power detection circuit 5 over a certain period of time for the sound output.

In the description of the embodiment of FIG. 1, as shown in the timing chart of FIG. 2, the ambient noise is stationary and there is almost no power change during the output of the warning sound. fluctuate. Therefore, if the threshold utterance circuit 6 is controlled only at the time point c and a new threshold value is set in proportion to the power output of the power detection circuit 5 at that time, an error may occur.

The averaging circuit 11 is controlled by the control circuit 9 and receives the power value output from the power detection circuit 5 during the audible sound output time.
This time average is output. Then, based on this time average value, the threshold value generating circuit 6 newly generates and holds a proportional threshold value.

As a result, in the present embodiment, even when the ambient noise level fluctuates slightly, the voice section can be detected accurately. Other operations are the same as those in FIG.

FIG. 4 is a block diagram for explaining a third embodiment of the present invention. In FIG. 5, the same reference numerals as in FIG. 1 denote the same parts.

In the figure, 12 is an echo canceling circuit, 121, 122, 123, 124,
Reference numeral 125 denotes an AD (analog digital) converter, DA (digital analog) converter, digital filter, adder / subtractor, tap coefficient setting circuit, and 104, which constitute the echo canceling circuit 12, respectively.

In the embodiment shown in FIGS. 1 and 3, a sine wave is output as a reminder sound, but in the present embodiment, a synthesized voice synthesized by the voice synthesis circuit 104 is output as a reminder sound.
This is because a human voice has a more natural and preferable impression as a reminder sound than a sine wave.

The voice synthesizing circuit 104 synthesizes voice using an electronic circuit based on data of characteristic parameters of a human voice created in advance. This synthesized speech is an instruction such as "Please ..." when asking someone for something. As a result, the operator can make a speech input to the speech recognition device in an interactive manner.

As is well known, speech is composed of many frequency components, and when this is used as a reminder sound, the reminder sound input to the microphone 2 is removed from the band such as BEF3 as in the embodiment of FIGS. It cannot be removed with a filter.

The sound emitted from the speaker 103 is reflected on the wall of the room or the like and is input to the microphone 2. Therefore, the output of the speaker and the input of the microphone have different phases for each frequency component, and the reminder sound input to the microphone 2 cannot be eliminated only by a simple phase shifter and an adder / subtractor. That is, it is impossible to measure the ambient noise level input to the microphone 2 during the utterance voice utterance.

The echo canceling circuit 12 includes an echo path, in this case, the speaker 10.
The transfer function of the path from 3 to the microphone 2 is estimated, and the speaker 103
Only the signal output from the microphone and input to the microphone 2 is deleted. The details of the echo canceling operation are described in, for example, "Digital Signal Processing" edited by the Institute of Electronics and Communication Engineers, and will be described only briefly here.

The echo canceling circuit 12 converts an analog signal into a digital signal, and internally performs signal processing by digital operation. The notification sound output from the speaker 103 is output from the AD converter 121.
Are input to the digital filter 123 and the tap coefficient setting circuit 125 as reference signals. Digital Filter 1
Reference numeral 23 denotes, for example, an FIR type filter, which is constituted by a delay line having a large number of tap outputs. Then, a copy (replica) of the notification sound as the reference signal is created, and the adder / subtractor 124 is generated.
Output to On the other hand, the signal input to the microphone 2 is input to the adder / subtractor 124 via the AD converter 121. In the adder / subtractor 124, a replica of the notification sound created from the reference signal is subtracted from the echo signal of the notification sound input to the microphone 2. As a result, at the output of the adder / subtractor 124, the reminder voice component is deleted and does not exist,
Other components (ambient noise) are output via the DA converter 122.

The tap coefficient setting circuit 125 sets the weight coefficient of the tap output of the digital filter 123, and the residual signal (output signal of the adder / subtractor 124) from which the reference signal and the echo signal have been eliminated.
The value is set from the correlation. As the setting algorithm, a known LMS method, a learning identification method, or the like is used. In brief, this is a method in which the tap weight coefficient is updated little by little so as to minimize the residual signal. When the residual signal becomes minimum, the weight coefficient value sequence set by the tap coefficient setting circuit 125 reaches the microphone 2 from the speaker 103 and is the best estimation of the impulse response as a time domain expression of the reverberation path transfer function.

The control circuit 9 controls the reverberation canceling circuit 12 during the time when the reminder sound is being output, so that the reverberation signal (input to the microphone 2) of the reminder sound is erased.

By the above erasing operation, the power detection circuit 5 is not disturbed by the notification sound component even during the output of the notification sound,
At the same time, the power of the ambient noise input to the microphone 2 is accurately output. Other operations are the same as those of the embodiment described with reference to FIG.

When the echo canceling circuit 12 is not performing the erasing operation, there is no output signal of the digital filter 123 (there is no input because there is no audible sound output), and the input signal to the microphone 2 is directly subjected to AD conversion and DA conversion. Power detection circuit 5
Is output to

The echo canceling circuit 12 requires a certain time (about 100 ms) to cancel the echo (estimating the echo path), so that the threshold value generating circuit 6 generates and holds a new threshold value (the first time). The time point c) in FIG. 3 is preferably after the erasure is completed (for example, after 100 ms or more has elapsed from the output of the audible alarm).

It is apparent that the averaging circuit 11 may be inserted before the threshold value generating circuit 6 as in the embodiment shown in FIG.
In this case, the averaging time is preferably the time from the completion of the erasure to the end of the output of the reminder sound.

In this embodiment, the notification sound is a synthesized voice.
It is clear that the echo canceling circuit 12 performs an erasing operation even if the audible tone is a sine wave as in the embodiment of FIGS.

As described above, according to the present embodiment, even if the reminder sound is a voice, the ambient noise level can be measured during this output, that is, immediately before the operator utters the voice. Thereby, one threshold value of the comparison circuit 7 for voice detection is set, and the operator voice is detected at the threshold value. Therefore, even when the ambient noise level fluctuates, the voice section can be reliably detected. Can be detected. And
It is possible to prevent the recognition rate of the speech recognition device from decreasing and malfunctioning.

FIG. 5 is a block diagram for explaining a fourth embodiment of the present invention, and the same reference numerals as those in FIG. 4 denote the same parts.

In the figure, reference numeral 13 denotes a noise microphone which is set apart from the microphone 2 so as to collect only ambient noise as much as possible, and 14 denotes a switching switch for switching a reference signal input of the echo canceling circuit 12. The switching switch 14 is a speaker 103
One of the audible sound output and the noise microphone 13 is supplied from the controller and supplied to the echo canceller circuit 12.

During the output of the notice voice, as in the embodiment shown in FIG.
The control circuit 9 controls the switching switch 14 and inputs the notification sound to the echo canceling circuit 12 as a reference signal. Then, the alarm sound input to the microphone 2 is deleted, and the power of the ambient noise can be detected by the power detection circuit 5 even while the alarm sound is being output. At times other than when the notification sound is being output, the control circuit 9 controls the switching switch 14 and the noise microphone 13
Is input to the echo canceller circuit 12 as a reference signal.

Here, as a reference signal input of the echo canceling circuit 12,
The operation when the output of the noise microphone 13 is selected will be described. At this time, the echo canceling circuit operates as a noise canceling circuit.

As described above, the digital filter 123 creates a replica from the ambient noise input to the noise microphone 13 and subtracts it from the ambient noise input to the microphone 2 to eliminate it. If the source of the ambient noise is sufficiently distant from the noise microphone 13 and the microphone 2 and the same ambient noise is input to both microphones, the signal output from the DA converter 122 includes the surrounding input to the microphone 2. No noise is output. However, in reality, the ambient noise input to the microphone 2 is partially output because it is slightly different. Furthermore, considering that inputting only ambient noise to the noise microphone 13 cannot spatially block sound waves, even if the distance between both microphones is separated, the noise microphone It is inevitable that the operator uttered voice is mixed. This causes disturbance to the ambient noise as a reference signal, and reduces the erasing ability.

However, even if there are some problems, it is possible to prevent the ambient noise from being mixed into the voice uttered by the operator after the output of the notification voice.

As described above, according to the present embodiment, the operator's uttered voice can be reliably detected as in the embodiment of FIG. In addition, the detected voice has a reduced amount of ambient noise mixed therein, and it is possible to prevent the recognition rate of the voice recognition device from lowering and prevent malfunction.

FIG. 6 is a block diagram for explaining a fifth embodiment of the present invention, in which the same reference numerals as those in FIG. 4 denote the same components, 15 is an adjusting circuit, and 16 is a differential amplifier circuit.

In the embodiments described above, the audible sound signal input to the microphone 2 is erased during the audible sound output, the surrounding sound level is detected, and an attempt is made to set the threshold value for the sound pressure detection based on the detected value. Things. In the case where the notification sound is a synthesized voice, a complicated echo canceling circuit 12 is required as in the embodiment of FIG.

In this embodiment, the same effect as that of the embodiment shown in FIG. 4 is obtained by a simpler method.

In FIG. 6, the notice sound output from the speaker 103 and the input sound to the microphone 2 are significantly different in phase and frequency characteristics due to the reverberation characteristics of the space, and are significantly different when compared by waveforms. But,
From the viewpoint of power, it is similar because energy conservation is established. The power time series of the audible sound emitted from the speaker 103 and the power time series of the audible sound input to the microphone 2 have different absolute values, but are similar in shape.

In the present embodiment, based on this principle, the ambient noise power input to the microphone 2 is detected during the utterance voice utterance, and the threshold value is set accordingly.

The power detection circuit 5 connected to the speaker 103 and the microphone 2 detects each power. The adjustment circuit 5 corrects the output power value (amplification or attenuation)
I do. Here, it is assumed that the positional relationship between the speaker 103 and the microphone 2 is fixed. Preferably, they are fixed to the same housing in advance. An alarm sound is output in the absence of ambient noise, and the output of the power detection circuit 5 connected to the microphone 2 and the output detected by the power detection circuit 5 connected to the speaker 103 and corrected by the adjustment circuit 15 are the same. Adjust the adjustment circuit so that

Output signals of the adjustment circuit 15 and the power detection circuit 5 connected to the microphone 2 are supplied to a differential amplifier circuit 16, respectively. Then, the differential amplifying circuit 16 supplies the difference signal between the two to the threshold value generating circuit 6. By the above adjustment, the output signal of the differential amplifier circuit 16 becomes zero. In other words, there is no ambient noise by the adjustment circuit 15 and only the audible signal is output and the microphone 2
Is adjusted so that the output of the differential amplifier circuit 16 becomes zero in a state where the signal is input to.

When the ambient noise is input to the microphone 2 in the state adjusted as described above, the power of the alarm sound output is canceled by the output signal of the differential amplifier circuit 16 even during the alarm sound output, and the power of the ambient noise is reduced. Only output. The control circuit 9 controls the threshold value generating circuit 6 based on the ambient noise power, and newly generates and holds a threshold value proportional to the power. The other operations are the same as those of the embodiment shown in FIG.

Note that the adjustment circuit 15 may be provided between the power detection circuit 5 connected to the microphone 2 and the differential amplifier circuit 16.

In some cases, it is preferable to provide a time delay function in consideration of the propagation delay of a sound wave from a speaker to a microphone in addition to the power value adjustment function.

Further, as in the embodiment of FIG. 3, the output signal of the differential amplifier circuit 16 may be supplied to an averaging circuit.

As described above, according to this embodiment, it is possible to provide an inexpensive and reliable voice detection output device with a simple circuit configuration as compared with the embodiment of FIG.

〔The invention's effect〕

As described above, according to the present invention, the ambient noise power can be accurately measured at the timing immediately before the operator's utterance, that is, at the time of outputting the audible sound. Therefore, even when the ambient noise level fluctuates, the voice section can be reliably detected. As a result, it is possible to prevent the recognition rate of the speech recognition device from being lowered and malfunctioning. In the case of a voice storage device, a reduction in file cost can be achieved.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a first embodiment of a voice detection and output device according to the present invention, FIG. 2 is a timing chart for explaining the operation of the configuration of FIG. 1, and FIG. 3 is a second embodiment of the present invention. FIG. 4 is a block diagram illustrating a third embodiment of the present invention, FIG. 5 is a block diagram illustrating a fourth embodiment of the present invention, and FIG. 6 is a block diagram illustrating the present invention. FIG. 7 is a block diagram for explaining a fifth embodiment of the present invention, and FIG. 7 is an explanatory diagram of the principle of detecting a voice section from an input signal. 1 ... notice sound output circuit 2 ... microphone 3 ... BE
F, 4 ... switch, 5 ... power detection circuit, 6 ... threshold value generation circuit, 7 ... comparison circuit, 8 ... switch, 11 ...
... Averaging circuit, 12 ... Echo canceling circuit, 13 ... Noise microphone, 14 ... Switching switch, 15 ... Adjustment circuit, 16 ...
Differential amplifier circuit.

Claims (9)

(57) [Claims] 1. A sound detection and output device for detecting and outputting a sound signal from an input signal from a microphone, wherein a sound signal is output to emit a sound to a space where the microphone is installed in order to notify a sound input. Means for erasing the audible sound component from the output signal of the microphone, and detecting the power of the output signal of the audible sound elimination means while the audible sound component is being output from the microphone. Power detecting means for detecting the power of the output signal of the microphone during a period other than the period; and generating and holding a threshold value corresponding to the power of the output signal of the audible tone erasing means detected by the power detecting means. Threshold value generation means; and comparison means for comparing the threshold value held by the threshold value generation means with the detected power of the power detection means. The threshold value generating means generates and holds the threshold value within the period of the sound signal component in the output signal of the microphone, and the sound after the sound signal component of the input signal from the microphone is output by the comparing means. An audio detection and output device for detecting and outputting an audio signal by detecting a section. 2. The voice detection output device according to claim 1, wherein the audible sound is a sine wave, and the audible sound elimination means is constituted by a band eliminator filter for removing a sine wave component. 3. The digital signal processing system according to claim 1, wherein the audible sound erasing means is a digital filter connected to the audible sound output means and comprising a delay line having a tap output; an adder / subtractor connected to the microphone; And a tap coefficient setting circuit for setting a weight coefficient of a line tap. 4. The digital microphone according to claim 3, further comprising a second microphone for collecting ambient noise, wherein the second microphone is connected to the digital filter during a period other than a period in which the audible sound is output from the audible sound output means. A voice detection output device. 5. The voice detection output device according to claim 1, further comprising averaging means for averaging the output of said power detection means with time. 6. A sound detection output device for detecting and outputting a sound signal from an input signal from a microphone, wherein a sound signal for emitting a sound to a space in which the microphone is installed to notify a sound input. Means, first power detection means for detecting the power of the output signal of the alarm sound output means, second power detection means for detecting the power of the output signal of the microphone, and the first and second powers A differential amplifying means for outputting an output difference of the detecting means, and a threshold for generating and holding a threshold corresponding to an output of the differential amplifying means during an output period of the alarm sound component from the microphone Generating means, and comparing means for comparing the output of the differential amplifying means with the threshold value, wherein the threshold value generating means operates within the period of the audible tone component in the output signal of the microphone. Generate and hold threshold , By detecting the speech interval after components 該催 Tsugeoto of the input signal from the microphone at said comparing means, voice detection output apparatus characterized by detecting and outputting an audio signal. 7. The sound detection output device according to claim 6, further comprising averaging means for averaging the output of the differential amplifying means over time. 8. A speech recognition device comprising the speech detection output device according to claim 1 as speech input means. 9. A voice storage device comprising the voice detection output device according to claim 1 as voice input means.