JPH09305196A - Speech input device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech input device to input a speech so that it is not heard by a third party in such a situation as the third party may hear it at an automatic cash dispenser, etc. SOLUTION: A video camera 11 reads a pattern of the eyes 2a of an utterer 2. An iris code generation part 12 extracts features of an iris pattern and generates iris data. The collation part 13 collates the iris data of the registered utterer in registration data base 20 to identify the utterer 2. A musical sound signal generation part 30 retrieves the speech data of the utterer 2 from the registered data base 20 and generates a musical sound signal to counteract the speech of the utterer 2 based on the speech data, to output it to a third party 3 from an acoustic output part 40. Thus, it becomes difficult for the third party 3 to recognize the speech of the utterer 2 by hearing. On the other hand, the speech of the utterer 2 is inputted to an acoustic input part 50 with the musical sound from the acoustic output part 40 synthesized, but the musical sound component is suppressed by a musical sound signal suppression part 60, and has no influence on the speech recognition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice input device used in an automatic depositing / dispensing device, etc., and particularly, in a situation where a third party may hear the voice, the voice is not heard by the third party. The present invention relates to a voice input device for inputting.

[0002]

2. Description of the Related Art Conventionally, a voice input device has been used as input means in a situation where it is difficult to manually input from a keyboard or the like, for example, for sorting work of packages, visual inspection of products and the like. For example, in the sorting work of the parcels, the destination name of the invoice attached to the parcels conveyed by the belt conveyor is viewed, and the area name of the destination is input by voice. The voice input device recognizes the input voice and outputs the corresponding area code to the sorting device. In addition, a system has been put into practical use, which uses a telephone that cannot be manually input originally and answers voice inquiries about the deposit balance via a telephone line.
Furthermore, with the progress of voice input technology, voice input is not limited to the special cases described above, but as a data input method that replaces the conventional manual input from a keyboard such as form input in an office computer or input to a word processor. It has been applied. For example, the automatic deposit / withdrawal device currently used in banks and the like uses a card for inputting an account number, and a manual input method of inputting a personal identification number from a keyboard or the like is used for identification. ing.
This method has a problem that a person who is not accustomed to the machine operation is troublesome in the input operation and cannot operate smoothly when complicated input is required. Voice input can be considered as a method for solving such a problem. In that case, the user will input his / her name, amount of money, etc. by voice in accordance with the guidance of the automatic deposit / withdrawal device installed in a bank that allows unspecified persons to enter / exit.

[0003]

However, in the conventional voice input device, the voice uttered in a noisy place,
It has been developed with a focus on how to input clear voice as voice transmitted via a transmission path of a limited frequency band, and protects the privacy that the input voice is heard by others. There was little consideration in terms of. Therefore, for example, when using voice input with an automatic depositing / dispensing device installed in a bank, it is possible that another person may ask for the utterance content, which causes a problem in terms of privacy protection. In order to solve this, it is costly to simply provide soundproof equipment around the speaker such as a soundproof wall, and noise is generated around the speaker to make it difficult for a third party to hear. That is, there was a problem in that the voice recognition rate was lowered. The present invention provides a voice input device which solves the problem of privacy protection as a problem that the above-mentioned conventional technique has.

[0004]

In order to solve the above-mentioned problems, a first aspect of the present invention includes, in a voice input device, a voice produced by a speaker at a predetermined position and a musical sound from the surroundings of the predetermined position. A sound input means for inputting a sound and converting it into an electric signal, and a musical tone signal generating means for generating a cancelable musical tone signal for a voice uttered by the utterer, mainly composed of a formant frequency component included in a vowel of the voice. And a sound emission range is set in the vicinity of the predetermined position and in a direction deviated from the sound input means, and the musical tone signal is converted into a sound and an average greater than the average level of the sound produced by the speaker. Musical sound output means for outputting the musical sound of power to the emission range of the sound, and subtracting the musical sound signal generated by the musical sound signal generation means from the output signal of the acoustic input means, and outputting the output signal of the acoustic input means. A musical tone signal suppressing means for suppressing the power of the musical tone signal below a threshold for voice recognition, and a voice detecting means for detecting as a voice signal when the power of the output signal of the musical tone signal suppressing means exceeds the threshold. Is equipped with.

According to the first aspect of the invention, since the voice input device is constructed as described above, the following operation is performed. The tone signal voicing means generates a tone signal such that the formant frequency component contained in the vowel of the voice is mainly used to cancel the voice uttered by the utterer. This musical sound signal is converted into a sound by the musical sound output means and is output with a power larger than that of the voice produced by the speaker. This makes it difficult for a third party near the speaker to hear the voice of the speaker. On the other hand, since the musical sound output from the musical sound output means is set so as to be emitted in a direction deviated from the acoustic input means, the musical sound input to the acoustic input means has a lower level than the voice produced by the speaker. Becomes a sound. The sound in which a low-level musical sound is superimposed on the voice of the speaker is converted into an electric signal by the sound input means. The musical tone signal component included in this electric signal is suppressed below the threshold for voice recognition by the musical tone signal suppressing means. The output signal obtained by the tone signal suppression means is
When the threshold value is exceeded, the voice detection means detects the voice signal. A second invention employs a speaker identification method using an iris as shown in FIG. 2 in order to identify a speaker who inputs a voice in a voice input device.

FIG. 2 is a structural diagram of the iris for explaining the principle of the present invention. The iris B of the eye A is a muscle that adjusts the opening degree of the pupil C in the peripheral portion of the pupil C of the black eye, and its pattern is composed of wrinkles D, an iris winding ring E, a fovea F, and the like. . The human eye A is almost formed by 6 months of pregnancy, at which point a hole opens in the pupil, and the opening (pupil)
It is known that the wrinkle D is chaotically generated outward. The growth of this cleft D stops by about the second year of life, and thereafter, it does not change for the rest of its life and forms a pattern peculiar to the person. The iris identification method for identifying an individual by paying attention to such characteristics of the iris B has the following features as compared with other identification methods such as fingerprint matching, signature matching, and personal identification number. (A) It is difficult to forge. (A) The recognition accuracy is high. (C) It can be recognized without contact. (D) There is no change with age.

According to a second aspect of the present invention, the voice input device according to the first aspect is provided with voice data corresponding to the formant frequency components of vowels in the voices of a plurality of registered speakers, and an iris B pattern of each registered speaker. Registration means in which the corresponding iris data is registered, and a pattern of the iris B of the speaker is detected at a predetermined position to extract the feature of the iris B, the extraction result and the iris data registered in the registration means. And match
A speaker identification unit for identifying which one of the plurality of registered speakers is the speaker is added. Further, the tone signal generating means is capable of retrieving the voice data of the speaker identified by the speaker identifying means from the registering means, and canceling the voice produced by the speaker based on the voice data. Changed to generate a tone signal. In the third invention, the tone signal generating means of the voice input device of the second invention is the most similar in frequency component to the voice data of the speaker, as compared with a storage unit for storing a plurality of tone data created in advance. And a selection unit for selecting the musical tone data and reading it from the storage unit. According to the second and third aspects of the invention, the speaker identification means detects the pattern of the iris B of the speaker and collates it with the iris data of a plurality of registered speakers registered in the registration means to identify the speaker. Is done. Further, the sound signal voicing means retrieves the voice data of the identified utterer from the registration means, and generates a tone signal for canceling the voice of the utterer based on the voice data. Then, with such a musical sound being output, the same voice input as in the first aspect of the invention is performed.

[0008]

1 is a block diagram of a voice input device showing an embodiment of the present invention. This voice input device is incorporated in an automatic depositing / dispensing device 1 of a bank, for example, and is used for inputting a name, an amount of money, etc.
It has 0. The speaker identification means 10 extracts a feature of the iris pattern from the video camera 11 which captures the image of the eye 2a of the speaker 2 standing at a predetermined position, and the feature of the iris pattern from the image of the eye 2a captured by the video camera 11. It is composed of an iris code generation unit 12 that generates an IR, and a collation unit 13 that collates a database based on the generated iris code IR to identify the speaker 2. The collation unit 13 has a registration means (for example, a registration database).
20 are connected. The registration database 20 includes voice data VD corresponding to the formant frequency components of vowels of a plurality of registered vocalists registered as voice input persons.
It is a database in which personal information such as the iris data ID of each registered speaker is registered.

The output side of the collating unit 13 is connected to a musical tone signal generating means (for example, a musical tone signal generating unit) 30. The tone signal generator 30 searches the registration database 20 for the voice data VD of the speaker 2 identified by the collator 13,
Based on the voice data VD, a tone signal MS having a masking effect of canceling the voice of the speaker 2 is generated. Musical sound signal generator 30
The output side of is a musical sound output means (for example, a musical sound output section) 40.
Connected to The musical tone output section 40 converts the musical tone signal MS into a musical tone MT and outputs the musical tone MT.
The radiation range 40a is directed toward the third party 3 who waits for a turn, for example, behind or on both sides of the speaker 2. Further, the magnitude of the output musical sound MT is set to have a power larger than that of the voice of the speaker 2.

On the other hand, this voice input device has a sound input means (for example, a sound input section) 50 for inputting a voice uttered by the speaker 2. The acoustic input unit 50 has a microphone 51 arranged at a position deviated from the sound emission range 40a of the musical sound output unit 40. The microphone 51 maximizes the sound produced by the speaker 2 who is in a predetermined position. In addition to inputting with sensitivity, the incoming musical sound MT or the like is input and converted into an electric signal. The output side of the sound input unit 50 is connected to a musical sound signal suppressing unit (for example, a musical sound signal suppressing unit) 60. The output side of the tone signal generator 30 is also connected to the tone signal suppressor 60. The musical tone signal suppressing unit 60 suppresses the musical tone signal component included in the output signal of the acoustic input unit 50 to a power that does not affect the voice recognition, and outputs the musical tone suppressing signal MR. The output side of the tone signal suppressing section 60 is connected to a voice detecting means (for example, a voice recognizing section) 70.
The voice recognition section 70 determines that the output signal of the tone signal suppression section 60 is a voice signal when the power of the output signal exceeds the threshold value TH, and further recognizes its content.

FIG. 3 is a block diagram showing the details of the tone signal generator 30, tone output unit 40, sound input unit 50 and tone signal suppressor 60 shown in FIG. The musical tone signal generation unit 30 includes a plurality of storage units (for example, memories) 31-1, 31-2, ..., 31-m that store musical tone signals MS mainly composed of different frequency components as musical tone data MD. Have Each of the memories 31-1 to 31-m includes a selection unit 32.
Connected to The selection unit 32 searches the registration database 20 based on the identification information of the speaker 2 given from the collation unit 13 of FIG. 1 and selects the musical sound data MD most similar to the voice data VD of the speaker 2. . Further, the selection unit 32 stores the selected musical sound data MD in the corresponding memory 3
1-i (i = 1 to m) is read and output. The output side of the selection unit 32 is connected to a digital / analog (hereinafter referred to as D / A) conversion unit 33. D /
The A converter 33 converts the musical sound data MD given as digital data into an analog musical sound signal MS. The output side of the D / A conversion unit 33 is connected to the musical sound output unit 40 and the musical sound signal suppression unit 60.

The tone output section 40 includes an amplifier 41 that amplifies the tone signal MS output from the D / A converter 33 to a required power, and a speaker 42 that converts the output of the amplifier 41 into a tone MT and emits the tone MT. It is configured. Sound input unit 50
Is composed of a microphone 51 arranged such that the direction of maximum sensitivity is substantially coincident with the direction of the voice emitted by the speaker 2, and an amplifier 52 for amplifying the electric signal obtained by the microphone 51 to a predetermined level. . Music signal suppression unit 60
Has a plurality of delay circuits 61-1, 61-2, ..., 61-n commonly connected to the output side of the tone signal vocalization section 30. The delay circuits 61-1 to 61-n include the speaker 42.
The musical sound MT output from the microphone 5 passes through the space
It is a circuit that generates a delay corresponding to the time to reach 1. These delay circuits 61-1 to 61-n are used for the musical tone MT.
Corresponds to various propagation paths caused by reflection on the surrounding walls, etc., so that different delay times are set. The output sides of the delay circuits 61-1 to 61-n are respectively connected to attenuators 62-1, 62-2, ..., 62-n. The attenuators 62-1 to 62-n include the delay circuit 6
The levels of the output signals 1-1 to 61-n are adjusted to the level of the musical tone signal MS obtained by the acoustic input unit 50 through the space. The output side of the attenuators 62-1 to 62-n is
It is connected to the input side of the adder 63. The adder 63 is
The instantaneous value of the input signal is added and output, and the output side thereof is connected to one input terminal of the subtractor 64. The output side of the acoustic input unit 50 is connected to the other input terminal of the subtractor 64. The subtractor 64 subtracts the output signal of the adder 63 from the output signal of the sound input unit 50 and outputs the result as the musical tone suppression signal MR. The output side of the subtractor 64 is connected to the voice recognition unit 70 of FIG.

Next, a method (1) of registering the voice data VD and iris data ID of the utterance registrant, which is performed prior to using the voice input device according to the embodiment of the present invention, and the present invention shown in FIGS. 1 and 3. The operation (2) of the voice input device according to the embodiment will be described. (1) Method of registering voice data and iris data FIG. 4 shows the registered speaker 4 in the registration database 20 of FIG.
5 is a configuration diagram showing an example of a registration device for previously registering personal information such as the voice data VD and the iris data ID of FIG. This registration device has a video camera 81 that catches the eyes 4a of the registered speaker 4. The video camera 81 converts the image data of the eyes 4a of the utterance registrant 4 into the iris data generation unit 8
2 is output. The iris data generation unit 82 extracts the iris feature from the input image data of the eye 4a and generates the iris data ID of the registered speaker 4. The output side of the iris data generation unit 82 is connected to the control unit 83 that controls the entire registration device. A guide message display unit 84 is connected to the control unit 83. The guide message display unit 84 displays a registration operation instruction for the registered speaker 4.

On the other hand, this registration device has a microphone 85 for inputting a voice uttered by the registered speaker 4 according to a guide message. The output side of the microphone 85 is connected to an analog / digital (hereinafter referred to as A / D) converter 87 via an amplifier 86. A /
The D converter 87 samples the input analog audio signal at a constant cycle and converts it into a digital signal. The output side of the A / D converter 87 is connected to the voice analysis unit 88. According to the guide message, the voice analysis unit 88 determines that the voice input by the registered speaker 4 is valid voice input when the power of the voice uttered by the registered voice speaker 4 is equal to or higher than a predetermined level and continues for a predetermined time or longer. Then, the voice analysis unit 88 performs processing such as formant frequency analysis included in the vowels of the input voice, and outputs the analysis result to the control unit 83 as voice data VD. The registration database 20 is connected to the control unit 83 via the communication network 90. The control unit 83 has a function of registering the personal information of the registered speaker 4 including the iris data ID generated by the iris data generation unit 82 and the voice data VD analyzed by the voice analysis unit 88 in the registration database 20. are doing.

Next, an example of a procedure for registering personal information using the registration device of FIG. 4 will be described. This registration device is installed in a place with little noise so that voice input can be performed smoothly. First, registered speaker 4
In accordance with the display of the guide message display unit 84, for example, a personal registration card and a personal identification number are input to an operation unit (not shown) for identity verification. When the control unit 83 confirms that the person is who he / she is from the input information, he / she registers the registered speaker 4 from a light source (not shown).
The eye 4a is irradiated with light. In conjunction with this, the video camera 81 captures the eyes 4a of the registered voice speaker 4, and the image of the eyes 4a is input to the iris data generation unit 82. The iris data generation unit 82 extracts iris features from the input image data of the eye 4a, generates the iris data ID of the registered speaker 4, and outputs the iris data ID to the control unit 83.

Next, the control unit 83 obtains the voice data VD of the registered speaker 4 in order to obtain the guide message display unit 84.
For example, a guide message such as "Please say" I love Oki Bank "" is displayed. The words to be uttered here are selected to include all vowels in order to accurately analyze the characteristics of the voice of the registered speaker 4. The uttered voice is converted into an electric signal by the microphone 85, amplified to a predetermined level by the amplifier 86, and converted into a digital signal by the A / D converter 87. The voice converted into a digital signal is input to the voice analysis unit 88, and is determined to be a valid input voice when the power of the voice is equal to or higher than a predetermined level and continues for a predetermined time or longer. If it is not determined that the input voice is valid, a guide message for prompting the registered speaker 4 to speak again is displayed, and the voice input is repeated. The effective voice input obtained in this way is
Processing such as formant frequency analysis is performed in the voice analysis unit 88, and the analysis result is output to the control unit 83 as voice data VD. Registered speaker 4 obtained in this way
The iris data ID and the voice data VD are registered in the registration database 20 from the control unit 83 via the communication network 90 together with other personal information of the registered speaker 4.

(2) Operation of the voice input device Now, when the speaker 2 stands at a predetermined position in front of the voice input device of FIG. 1, a sensor (not shown) detects the speaker 2 and is shown. The eyes 2a of the speaker 2 are illuminated with light from a light source that is not present. In conjunction with this, the video camera 11 captures the eye 2a of the speaker 2, and the image of the eye 2a is input to the iris code generation unit 12. Iris code generator 1
2 produces an iris code IR that is characteristic of the iris of the speaker 2. The iris code IR is collated by the collating unit 13 with the iris data ID of the registered speaker registered in the registration database 20 to identify who the speaker 2 is. The identification result of the speaker 2 identified by the collation unit 13 is provided to the selection unit 32 of the tone signal generation unit 30. The registration database 20 is searched by the selection unit 32, and the voice data VD of the creator 2 is read. Further, in the selection unit 32, the contents of the memory 31-i storing the tone data MD most similar to the voice data VD are read out and output to the D / A conversion unit 33. The musical sound MT obtained by being converted into analog by the D / A conversion unit 33 is amplified by the amplifier 41 of the musical sound output unit 40, and the third party 3 or the like waiting in turn behind the speaker 2 from the speaker 42. Is output toward. The volume of the musical sound MT output to the third party 3 or the like is adjusted to be larger than the volume of the voice produced by the speaker 2. Furthermore, since the frequency component of the musical sound MT and the formant frequency component of the voice of the speaker 2 are similar, it is difficult for the third party 3 to distinguish the voice of the speaker 2. In this state, the speaker 2
According to an instruction from a guidance output unit (not shown), a voice is input to the microphone 51 of the sound input unit 50. To the microphone 51, in addition to the voice produced by the speaker 2, the musical tone MT output from the speaker 42 is also circulated and input. These input sounds are converted into electric signals by the microphone 51, amplified by the amplifier 52 to a predetermined level, and given to the subtraction unit 64 of the musical sound signal suppression unit 60.

On the other hand, the output signal of the tone signal vocalization section 30 is
It is given to the delay circuits 61-1 to 61-n of the musical sound signal suppressing unit 60, and is equivalent to the propagation time by the plurality of acoustic propagation paths from the speaker 42 to the microphone 51 by each delay circuit 61-1 to 61-n. The amount of delay to do is given. Furthermore, the attenuators 62-1 to 62-n provide attenuation amounts corresponding to the attenuation in each acoustic propagation path. The output signals of the respective attenuators 62-1 to 62-n are added by the adder 63 and given to the subtractor 64. Subtractor 6
In 4, the signal given by the adder 63 is subtracted from the signal given by the sound input unit 50, and the result of the subtraction is output as the musical tone suppression signal MR. Since the positional relationship between the speaker 42 and the microphone 51 is determined when the voice input device is installed, the sound propagation path from the speaker 42 to the microphone 51 is almost constant. Therefore,
Before operating the voice input device, the delay circuits 61-1 to 61-n
And the attenuation amounts of the attenuators 62-1 to 62-n are adjusted, and the tone signal M included in the tone suppression signal MR is
It is assumed that the power of S is set to be smaller than the threshold TH for voice recognition. The musical tone signal suppressing unit 60 suppresses the musical tone signal component below the threshold value TH, and the musical tone suppressing signal MR becomes almost only the voice of the speaker 2.
Is output to the voice recognition unit 70. The voice recognition unit 70 detects a voice signal by the threshold TH for voice recognition and performs voice recognition.

In this way, the voice input device of this embodiment is
The speaker 2 is identified by the speaker identification means 10 by iris identification, and the tone signal voicing unit 30 retrieves the voice data VD of the speaker 2 from the registration database 20 to obtain a tone signal most similar to the voice of the speaker 2. Output MS. And
Since the musical sound MT is output from the musical sound output unit 40 to the third party 3, it becomes difficult for the third party 3 to hear the voice of the utterer 2 because the voice MT is masked by the musical sound MT. Further, the musical sound signal suppression unit 60 suppresses the musical sound MT which is sneaked from the musical sound signal output unit 40 and input to the acoustic input unit 50 by using the musical sound signal MS output from the musical sound signal vocalization unit 30.
By using the musical sound MT in this way, it becomes possible to protect privacy in the voice input of the speaker 2 without affecting voice recognition. In addition, since the identity verification is performed by the iris identification, there is an advantage particularly in the application side where security is required to be secured like the automatic depositing / dispensing apparatus of the present embodiment.

The present invention is not limited to the above embodiment, and various modifications can be made. For example, there are the following modifications. (A) It is possible to omit the speaker identification means 10 and the registration database 20. In that case, personal information such as the voice data VD of the speaker 2 cannot be obtained, so that the musical tone signal vocalization unit 30 assumes the standard speaker, not the individual speaker 2, and provides a masking effect for the voice. It is sufficient to generate a certain tone signal MS. With this configuration, the privacy protection effect may be slightly reduced, but the configuration of the device can be simplified. (B) In the above embodiment and the voice input device of the above (a), the example of being incorporated in the automatic depositing / dispensing device 1 has been described.
The invention can be similarly applied to a device requiring privacy protection, such as a betting ticket purchase terminal using voice recognition. (C) The tone signal generator 30 uses the memories 31-1 to 31-m in which various tone data are stored in advance, and selects a tone that is most similar to the voice of the speaker 2. The musical sound MT may be synthesized and output based on the audio data VD. This makes it possible to generate a musical tone MT having a greater masking effect. (D) Although the registration database 20 of FIG. 1 has a configuration incorporated in the voice input device, it may have a configuration of being accessed via a communication line network. This makes it possible to manage the personal information of the iris data ID and the voice data VD in one place, and thus it is possible to configure a system having expandability and flexibility.

[0021]

As described in detail above, according to the first aspect of the invention, the tone signal generating means and the tone outputting means provide
A musical sound that cancels the voice of the speaker is output. This makes it difficult for a third party near the speaker to hear the voice of the speaker. On the other hand, since this musical sound is set to be emitted in a direction deviated from the sound input means, it is input to this sound input means at a level lower than that of the voice produced by the speaker. Further, the musical tone signal suppressing unit suppresses the musical tone signal component to a value equal to or less than the threshold for voice recognition. Therefore, without affecting the voice recognition,
A voice input device capable of protecting privacy can be provided to a third party. According to the second and third aspects, the speaker is identified by the speaker identifying means. Further, the sound signal voicing means retrieves the voice data of the identified utterer from the registration means, and generates a tone signal for canceling the voice of the utterer based on the voice data. For this reason, it is possible to output a musical sound having an effect of canceling the voice uttered by the speaker more than in the first aspect of the invention, and thus it is possible to provide a voice input device having a further privacy protection effect.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a voice input device showing an embodiment of the present invention.

FIG. 2 is a structural diagram of an iris for explaining the principle of the present invention.

FIG. 3 is a diagram showing a musical tone signal generator 30, a musical tone output unit 40, and
3 is a configuration diagram of a sound input unit 50 and a musical sound signal suppression unit 60. FIG.

FIG. 4 is a configuration diagram of a registration device for the registration database 20 according to the embodiment of this invention.

[Explanation of symbols]

 2 Speaker 10 Speaker identification means 20 Registration database 30 Musical sound signal generation part 31-1 to 31-m Memory 32 Selection part 40 Musical sound output part 50 Sound input part 60 Musical sound signal suppression part 70 Speech recognition part A Eye B Iris C Pupil

Claims (3)

[Claims] 1. A sound input means for inputting a voice uttered by a speaker at a predetermined position and a sound including a musical sound from the surroundings of the predetermined position and converting the sound into an electric signal, which is included in a vowel of a voice. A musical sound signal generating means mainly for formant frequency component, which generates a musical sound signal that can be canceled with respect to the voice uttered by the speaker, and a sound signal in the vicinity of the predetermined position and in a direction deviated from the sound input means. A sound output means for setting a radiation range, converting the musical sound signal into a sound, and outputting the musical sound having an average power larger than the average level of the voice produced by the speaker to the radial range of the sound, and the acoustic input. A musical tone signal suppressing means for subtracting the musical tone signal generated by the musical tone signal generating means from the output signal of the means, and suppressing the power of the musical tone signal in the output signal of the acoustic input means below a threshold for voice recognition. A voice input device comprising: a voice detection unit that detects a voice signal when the power of the output signal of the tone signal suppression unit exceeds the threshold value. 2. A registration unit in which voice data corresponding to a formant frequency component of a vowel in voices of a plurality of registered speakers and iris data corresponding to an iris pattern of each registered speaker are registered, and a predetermined registration unit. The iris pattern of the speaker is detected at the position, the features of the iris are extracted, the extraction result is compared with the iris data registered in the registration means, and the speaker is selected from among the plurality of registered speakers. The speaker identification means for identifying which one of the corresponding persons, the voice uttered by the speaker at the predetermined position, and the sound including the musical sound from the vicinity of the predetermined position are input as an electric signal. A sound input means for converting the voice data of the speaker identified by the speaker identification means into the registration means, and the voice generated by the speaker can be canceled based on the voice data. Music A tone signal generating means for generating a signal, and a sound emission range is set in the vicinity of the predetermined position and in a direction deviated from the sound input means, and the tone signal is converted into a sound and emitted by the speaker. The musical sound having an average power larger than the average level of the voice, a musical sound output means for outputting to the emission range of the sound, and subtracting the musical sound signal generated by the musical sound signal generating means from the output signal of the acoustic input means, A musical tone signal suppressing means for suppressing the power of the musical tone signal in the output signal of the acoustic input means to a threshold for voice recognition or less, and a voice signal when the power of the output signal of the musical tone signal suppressing means exceeds the threshold value. An audio input device, comprising: 3. The musical tone signal generating means selects the musical tone data having a frequency component most similar to the voice data of the speaker and a storage unit for storing a plurality of musical tone data created in advance. The voice input device according to claim 2, further comprising a selection unit for reading from the unit.