JP2000222000A - Voice recognition device - Google Patents

Abstract

(57) [Summary] [Problem] To collect environmental noise at appropriate timing,
Provided is a speech recognition device that can obtain accurate environmental noise. SOLUTION: A microphone, an operation member for operating the microphone, first switch means which is turned on by a first stroke of the operation member, and is turned on by a second stroke following the first stroke of the operation member. A second switch, and a signal component input to the microphone when the first switch is turned on is subtracted from a signal component input to the microphone when the second switch is turned on. A computing unit; and a speech recognition unit that performs a speech recognition operation using an output of the computing unit.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition device for recognizing speech.

[0002]

2. Description of the Related Art In recent years, portable small precision devices and communication devices such as cameras have been highly digitized, and it has become possible to provide a great number of functions in spite of their small body size. . However, a large number of operating members such as electronic dials, push buttons, and slide switches are used to operate these functions, which makes it difficult to understand the operating method, and the number of operating members that can be arranged in a limited size of equipment. Due to limitations, sometimes complicated operations are complicated, such as pressing a plurality of operation members at the same time or operating sequentially in a hierarchical manner. For example, in the case of a camera that is a representative of small precision equipment, the conventional operation method is not only complicated and troublesome, but also it is difficult to operate while holding the camera under shooting conditions that require quickness. There was a problem in both the quick shooting performance.

In order to solve the above problems, Japanese Patent Laid-Open Publication No. Sho 64-56428 discloses a control mechanism for controlling the function of a camera, in which a voice input means for inputting voice and a voice for recognizing the input voice are provided. There has been proposed a voice input camera including a recognition unit and a control unit that controls a function of the camera based on control content corresponding to a recognition result. Accordingly, it is an object of the present invention to provide a camera excellent in operability and continuous shooting in which functions of the camera such as an aperture, a shutter speed, and an operation mode can be freely set by voice.

[0004]

SUMMARY OF THE INVENTION
Portable small precision equipment and communication equipment having a voice input function disclosed in Japanese Patent Publication No. 64-56428 (hereinafter abbreviated as voice input equipment)
Although the operation is simple, it is required that the degree of recognition is accurate. Therefore, how accurately the voice can be captured in the operation of the device and the recognition can be accurately performed realizes the voice input device. The above was a challenge. In particular, since these voice input devices are used in various environments both indoors and outdoors, it has been the most important issue to accurately recognize the voice of the operator even in each noise environment.

[0005] In addition, some recognition methods have been proposed in the past, which have improved recognition rates even in a noisy environment. For example, a region where a voice is generated by an operator and a region where the voice is not generated are separated by a threshold of a certain voice level, a non-voice region is set as a noise component, and this noise component is subtracted from a voice component uttered by the operator. A voice recognition method has been adopted in which a feature amount is obtained, and a matching process is performed with a pattern registered in advance. However, it was not possible to accurately separate only the noise component, resulting in inaccurate speech recognition, and it was necessary to accurately extract only the noise without making the operator aware.

Further, a voice input system having a microphone for collecting environmental sounds in addition to a voice input microphone has been proposed. However, in a portable small precision device or a communication device, the shape becomes large and it is not practical. Absent.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and a first object of the present invention is to provide a speech recognition apparatus capable of accurately collecting environmental noise by collecting environmental noise at an appropriate timing. It is in.

It is a second object of the present invention to provide a speech recognition apparatus which determines whether or not it is necessary for the speech recognition apparatus itself to collect environmental noise and, if necessary, automatically collects environmental noise.

[0009]

According to a first aspect of the present invention, there is provided a microphone, an operation member for operating the microphone, first switch means for turning on the first stroke of the operation member, Second switch means that is turned on in a second stroke following the first stroke of the operating member; and the first component is obtained from a signal component input to the microphone when the second switch means is turned on.
And a voice recognition means for performing a voice recognition operation using an output of the calculation means when the switch means is turned on.

According to a second aspect of the present invention, a microphone, first switch means for operating the microphone when a user's eye is at a predetermined position, and the user operates a predetermined operation member In this case, the second switch means for operating the microphone, and the second switch means when the first switch means is turned on based on a signal component inputted to the microphone when the second switch means is turned on. It is characterized by having arithmetic means for subtracting a signal component inputted to the microphone, and speech recognition means for executing a speech recognition operation using the output of the arithmetic means.

According to a third aspect of the present invention, there is provided a microphone, and first switch means for operating the microphone when a part of a user's body is at a predetermined position;
A second switch for operating the microphone when the user operates a predetermined operation member; and a first component based on a signal component input to the microphone when the second switch is turned on. And a voice recognition means for performing a voice recognition operation using an output of the calculation means when the switch means is turned on.

According to a fourth aspect of the present invention, a microphone, first switch means for operating the microphone, second switch means for operating the microphone, and when the second switch means is turned on. From the signal component input to the microphone to the first
And a voice recognition means for performing a voice recognition operation using an output of the calculation means when the switch means is turned on.

According to a fifth aspect of the present invention, in the voice recognition apparatus for detecting an environmental noise signal and performing a voice recognition operation using a signal obtained by subtracting the environmental noise signal from an input voice signal, And an environmental noise detecting means for detecting an environmental noise signal when an image captured by the image capturing means changes.

According to a sixth aspect of the present invention, in the voice recognition apparatus for detecting an environmental noise signal and performing a voice recognition operation using a signal obtained by subtracting the environmental noise signal from an input voice signal, And an environmental noise detecting means for detecting an environmental noise signal when the luminance of the image captured by the image capturing means changes by a predetermined value or more.

According to a seventh aspect of the present invention, in the voice recognition apparatus for detecting an environmental noise signal and performing a voice recognition operation using a signal obtained by subtracting the environmental noise signal from an input voice signal, And an environmental noise detecting means for detecting an environmental noise signal when a defocus amount of an image captured by the image capturing means changes by a predetermined value or more. .

According to the present invention, there is provided a speech recognition apparatus for performing a speech recognition operation by detecting an environmental noise signal and comparing a signal obtained by subtracting the environmental noise signal from an input audio signal with a reference audio pattern. In the case where the reliability determination means for determining the reliability of the speech recognition degree based on the difference between the signal and the reference pattern, and the reliability determination means determines that the reliability of the speech recognition degree is low ,
Environmental noise detection means for detecting an environmental noise signal.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) Hereinafter, embodiments of the present invention will be described in detail.

FIGS. 1A, 1B and 1C show a single-lens reflex camera having an audio input function showing an embodiment when the present invention is applied to a single-lens reflex camera which is one of portable small precision machines. It is a schematic diagram of the upper surface, the back surface, and the side surface of the camera.

In FIG. 1, 1 is a camera body, 2 is a release button, 3 is an AE mode setting button for setting an AE mode such as a well-known program AE, shutter priority AE, aperture priority AE, depth of field priority AE, etc. 4 is a known one-shot AF, AF mode setting button for setting an AF operation mode such as servo AF, 5 is a known evaluative metering, average metering, partial metering, spot metering, etc. Reference numeral 6 denotes an input switch generally called an electronic dial, which generates two click pulses having different timings when rotated.
When the setting buttons 3 to 5 are pressed to enter the mode setting state, each mode is sequentially displayed and selected on a monitor LCD described later.

Reference numeral 9 denotes a monitor LCD as an external monitor display device, which comprises a fixed segment display section 9a for displaying a predetermined pattern and a 7-segment display section 9b for displaying a variable numerical value. Reference numeral 10 denotes a camera back cover having a voice recognition unit which is the center of the configuration of the present embodiment.

Reference numeral 11 denotes a voice input button which is a trigger switch for inputting a voice uttered by the photographer.
The sub electronic dial 12 used for setting the number of exposure correction steps when performing AE photographing also provided on the back lid 10 in the same configuration as
Are provided at the center of rotation.

FIG. 2 is a local sectional view showing the configuration of the switch section of the voice input button 11. As shown in FIG. This switch is a known two-stage switch that has two strokes in the pressing direction and turns on and off two circuits. When the key top 11a is pressed, the outer peripheral portion 22b of the bowl-shaped contact spring 21 thereunder is first dented by the first stroke, and the GND pattern 22c provided on the switch board 22 and the SW-NOISE pattern 22b described later are conducted. And SW-NOISE turns on.

When the key top 11a is further depressed, the central portion 22a of the contact spring 21 is depressed by the second stroke, and the GND pattern 22c provided on the switch board 22 and the SW-VO to be described later are pressed.
The pattern 2a of the ICE is made conductive, and the SW-VOICE is turned on. The spring pressure of the first stroke is 5-7 for the second stroke.
It is appropriate that the latter is made about twice as heavy and the first stroke is set to about several tens g, which is turned on only by putting a finger lightly. In other words, SW-NOISE turns on unconsciously,
SW-VOICE does not turn on unless you push it quite consciously.

Returning to FIG. 1, reference numeral 13 denotes a voice mode for selecting three positions: a position for turning off a voice input function, a voice recognition mode for performing a voice recognition operation, and a voice registration mode for pre-registering a voice of a photographer. A switch 14 is a small micro speaker configured to generate sound from a hole formed in the back cover 10, and 15 is a small electret condenser microphone that captures a photographer's voice.

FIG. 3 is a block diagram showing an electrical configuration incorporated in the single-lens reflex camera having the above-described configuration. The same components as those in FIG. 1 are denoted by the same reference numerals. In the drawing, a block diagram surrounded by a two-dot chain (A) shows a camera function unit built in the camera body 1, and a block diagram surrounded by a two-dot chain line (B) shows a voice built in the back cover 10. 4 shows a recognition unit. First two
The configuration in the block diagram showing the camera function unit built in the camera body 1 and surrounded by a chain line will be described.

A central processing unit (hereinafter, referred to as a main CPU) 101, which is a microcomputer built in the camera body 1, includes an automatic focus detection circuit 102, a focus adjustment circuit 103,
Photometry circuit 104, shutter control circuit 105, aperture control circuit 10
6. The motor control circuit 107 is connected. This main
When the first stroke of the release button 2 is pressed, the CPU 101 detects a focus state of a photographic lens (not shown) and performs a so-called AF operation for driving a focus adjustment mechanism of the photographic lens based on the state. The brightness of the subject to be measured is measured, and the exposure value is determined based on the measured light value.

Next, by further pressing the release button 2 to the second stroke, the shutter and the aperture of the photographing lens are controlled at a predetermined shutter time and an aperture value, and the film is exposed at an exposure amount corresponding to the exposure value. After the exposure is completed, the film is wound up by one frame and the shutter is charged, thereby executing a series of camera release sequences.

SW-1 is a switch that is turned on by the first stroke of the release button 2 to start AF and photometry, and SW-2 is a release switch that is turned on by the second stroke of the release button 2. SW-AEMD is a switch linked to AE mode setting button 3, SW
-AFMD is a switch linked to AF mode setting button 4, SW-MEMD
Is a switch linked to a switch linked to the photometry mode setting button 5.

SW-DIAL1 and SW-DIAL2 are dial switches provided in the electronic dial 8, and the signal input circuit 107
Is input to the up / down counter of the electronic dial 8 to count the amount of rotation click of the electronic dial 8. The above switch states are input to the signal input circuit 107 and transmitted to the main CPU 101 via the data bus. Reference numeral 108 denotes an LCD drive circuit having a known configuration for driving the LCD to display an aperture value, a shutter time, a shooting mode, the number of films, and the like on the monitor LCD 9 in accordance with a signal from the main CPU 101. At the time of second, it is also displayed on the LCD 109 in the viewfinder.

Reference numeral 110 denotes a microprocessor that mainly performs voice recognition processing. A voice signal output from the microphone 15 is input to a preamplifier 111, amplified at a predetermined gain, sent to an A / D converter 113, and converted into a digital voice signal. The data is converted and sent to the microprocessor 110, where voice recognition processing is performed. The result of the voice recognition and the voice recognition operation status are transmitted to the main CPU 101 via the data bus. Note that the microprocessor 110 causes the gain control 112 to perform feedback control, that is, so-called automatic gain control (AGC) so that a volume suitable for voice recognition is input.

Reference numeral 114 denotes a RAM provided as a working memory for previously storing a photographer's voice and acoustic feature parameters of environmental noise, and as a working memory for performing voice recognition processing.
Reference numeral 15 denotes a ROM in which voice data to be uttered from the camera is stored in advance, and both are connected to the microprocessor 110 via the memory controller 116. 117
Is a D / A converter. The microprocessor 110 calls up audio data stored in the ROM 115 via the memory controller 116, and converts the audio data into an analog audio signal. The sound is further amplified by the power amplifier 118 so as to have an appropriate volume, and the sound stored from the speaker 14 is uttered. SW-VMD is a three-position switch that works with the audio mode switch 13. SW-NOISE is a noise input switch that is turned on by the first stroke of the voice input button 11, and SW-VOICE is a voice input switch that is turned on by the second stroke of the voice input button 11.

In general, speech recognition devices are classified into those for specific speakers that limit the speakers and those for unspecified speakers that can recognize any voice without limiting the speakers. For a specific speaker, since a recognition system can be set for a specific speaker to be used, the load on the system can be reduced, a high recognition rate can be expected, and the language is less dependent on language.

However, the absolute inconvenience of forcing the user to utter a vocabulary to be recognized and register it in advance is inevitable. On the other hand, for unspecified speakers, it is easy to operate voice recognition immediately, regardless of the speaker.However, in order to improve recognition accuracy, a large-scale system is required for both the arithmetic unit and the memory. come.

By the way, from the point of view of an application such as a camera and a portable small precision device, there are not so many functions for which voice input is desired (approximately 100 vocabulary), and in most cases the user is limited to one individual. Considering the above characteristics and the absolute condition that small size and low cost are absolute conditions, it can be said that a speech recognition device that is a specific speaker and targets a specific vocabulary is suitable. Against this background, the features of the apparatus having the voice input function according to the present invention are also suitable for the specific speaker specification.

FIG. 4 is a block diagram of a speech recognition process performed by a speech recognition device centered on the microprocessor 110. When the digitized audio signal is input to the microprocessor 110, the spectrum analysis unit 201 performs frame processing to cut out a certain section in a time series in order to detect a feature amount of the audio, and performs spectrum analysis by Fourier transform. Find the input spectrum. Noise removal unit 202
Performs a so-called spectrum subtraction process of reading a noise spectrum stored in the noise pattern recording unit 203 in the RAM 114 in advance and obtaining a speech spectrum to be recognized by subtracting the noise spectrum from the input spectrum.

Many studies have been published on the algorithm of the spectral subtraction method, including Boll, IEEE Trans. Vol. Assp-27, No. 2, April 1979.

The feature quantity extraction unit 204 calculates the feature quantity of the input speech spectrum in frame units. This is configured to extract speech feature vectors such as speech power, linear prediction coefficients (LPC), and cepstrum coefficients for each predetermined band.

The reference voice pattern storage unit 206 in the RAM 114 stores a reference voice pattern (feature vector sequence) registered in advance by extracting a feature vector by the same voice analysis system. A matching process is performed between the reference voice pattern and the feature vector of the voice to be recognized. The matching calculation is performed as a distance calculation between the reference voice pattern vector and the voice pattern vector to be recognized.

This calculation algorithm is based on the feature amount of the feature extraction unit, for example, DP (dynamic programming).
The matching process is performed according to a predetermined speech recognition algorithm such as a matching method or an HMM (Hidden Markov) method. The next determination unit 207 determines that the word with the smallest distance from each registered reference voice pattern is recognized as a recognized word, and outputs the word as a recognition result.

Next, a specific operation of the speech recognition apparatus in consideration of environmental noise which is a feature of the present invention will be described. FIG. 5 shows that the voice mode switch 13 is in the registered position,
9 is a flowchart illustrating an operation in a “registration mode” in which a photographer's voice is registered in advance.

When the voice mode switch 13 is at the registration position and the VMD-SW is ON at the registration side, the process enters the "registration mode" at 301. Then, whether any of the mode setting buttons in 302 is pressed, that is, AEMD-SW, AFMD-SW, MEM
Detects whether D-SW is ON. Both switches are OF
If F is performed, this detection is repeated until it is turned ON. If any of them is ON, the process proceeds to 303, and the mode timer starts. Next, at 304, display the mode setting status for monitor L
The desired mode can be selected by displaying the image on the CD 9 and rotating the electronic dial 8 by the photographer (30).
Five).

An example of this state will be described with reference to FIG. FIG. 6 shows a display state on the fixed display unit 9a when the photometry mode setting button 5 is pressed. As shown in FIG. 6, when the electronic dial 8 is rotated clockwise and counterclockwise, evaluation photometry, partial photometry, spot photometry,
Average photometry can be sequentially selected and the photometry mode can be set. AE
The same can be set in the mode setting and the AF mode setting.

When any one of the photographing modes is selected, the microprocessor 11 determines whether the first stroke of the voice input button 11 is pressed at 306 and the SW-NOISE is turned on.
0 is detected. If it is OFF, the process proceeds to step 307. If the mode timer has passed a predetermined time, the process returns to step 302. If not, the process returns to 304 and the mode setting display is continued.

If SW-NOISE is ON at 306, the microprocessor 110 proceeds to 308 to start noise detection processing. That is, the sound after this is taken in from the microphone 15 as environmental noise. Next, the processing proceeds to noise analysis 309, in which the spectrum analysis unit 201 of the microprocessor 101 performs frame processing for cutting out a certain section of the digital audio signal, and performs spectrum analysis by Fourier transform to obtain a noise spectrum for a predetermined time.

Next, the process proceeds to 310, where the microprocessor 110
Estimates a noise spectrum to be stored from the obtained noise spectrum. This is estimated, for example, by calculating an average spectrum of spectra for several frames. next
Proceeding to 311, the estimated noise spectrum is stored in the noise pattern recording unit 203 via the memory controller 116.
It is stored in the RAM 114.

Next, at 312, it is detected whether the second stroke of the voice input button 11 is pressed and the SW-VOICE is turned on. OFF
If so, the process returns to 306, and if it has been turned on, the microprocessor 110 ends the noise detection process at 313. Therefore 3
This means that environmental noise was acquired between 06 and 313. The noise pattern is repeatedly detected after a lapse of a predetermined time, but is always updated to the latest noise pattern and stored. Upon completion of the noise detection, the microprocessor 110 reads the settings of the camera mode from the main CPU 101, and starts detecting the input voice of the photographer at 314. That is, the sound after this is taken in from the microphone 15 as the voice of the photographer.

Here, the photographer utters the mode name so that the input voice to be recognized as the selected mode is registered in correspondence with the selected mode. For example, evaluation photometry mode Fig. 6 (a)
Is selected, and if the partial photometry mode is selected in FIG. Then, the microprocessor 110 proceeds to a voice spectrum analysis 315 of the input voice, and the spectrum analyzer 201
A frame processing for cutting out a predetermined section is performed on the digital audio signal captured in step (1), and spectrum analysis is performed by Fourier transform to obtain an input spectrum.

Next, proceeding to 316, the microprocessor 110
The noise elimination unit 202 reads the noise spectrum stored in the RAM 114 of the noise pattern recording unit 203 with respect to this input spectrum and subtracts the noise spectrum from the input spectrum to obtain a speech spectrum to be recognized. I do.

An example of the noise removal operation will be described with reference to FIG. 7 (1) to 7 (3) show the relationship between the frequency and the spectrum at a certain time, the speech input from the microphone, the noise stored in advance, and the speech after the noise removal operation, that is, the speech recognition processing is performed. It is an expression of the sound to be played.

The spectrum pattern S1 of the input voice at a certain time: S11, S12,... S1n indicates the frequency spectrum at each frequency.
N: N1, N2,... Nn are also shown in the same manner as the spectrum pattern of the input voice. S1p, S2p, S3p... Indicate the power spectrum in each frequency band of the input voice on each time axis, and Np indicates the power spectrum in each noise frequency band. Here, the result after the noise removal operation on the time axis S1 is S1i ′, and the spectral pattern of the input voice is S1i ′.
i, assuming that the noise spectrum pattern is Ni, the noise removal operation is expressed as the following equation.

S1i ′ = S1i− (S1p / Np) * Ni ··· (1) That is, the calculation result is estimated by multiplying the noise included in the input voice by the ratio of the power spectrum. It is the result of subtraction.

Next, the routine proceeds to 317, where the microprocessor 110
Is calculated by the feature amount extraction unit 204 for each frame in the calculated speech spectrum pattern. This includes speech power and linear prediction coefficients (LP
C), speech feature vectors such as cepstrum coefficients are extracted. For example, a linear prediction coefficient is obtained by performing a linear prediction analysis process, and a cepstrum coefficient (LPC cepstrum) is calculated from the linear prediction coefficient. As a result, a voice pattern to be registered is generated.

Next, the flow proceeds to 318, where the reliability of the voice pattern is determined. That is, it is determined whether the generated voice pattern has reached a level worth registering as a reference pattern. If the reliability is determined to be insufficient 31
Proceeding to 9, the registration is disabled, and a display recommending re-entry is made to perform the registration operation again. This flashes the mode display section to be set, which is displayed on the LCD 9 for the monitor, and "Cannot register" from the speaker 14.
Again "to inform the photographer.

Then, this recommendation display is performed for a predetermined time,
After resetting the mode timer, the process returns to 306, and waits for the voice input SW1 to be pressed again. If the reliability is determined to be OK, the process proceeds to 320, and it is checked whether the number of voice patterns made so far has reached a predetermined number n.
Then, as in 319, the speaker 14 recommends “again” by voice. Reset mode timer after advisory 30
Return to 6. If the number has reached the predetermined number n, the process proceeds to 322 to create a reference voice pattern to be registered. This is created from any one of the average value and intermediate value of the n voice patterns, the voice pattern with the highest reliability, and the like. Next, the process proceeds to 323, where the reference voice pattern is stored in the reference voice pattern recording unit 206 of the RAM 14, and the registration operation is completed.

Next, a description will be given of a "voice recognition mode" in which a voice is actually input to the camera. FIG. 8 is a flowchart for explaining the operation. The microprocessor 110 detects the state of the voice mode switch 13, and notifies the main CPU 101 that the voice mode switch 13 is in the recognition position and that the VMD-SW is ON on the recognition side and that 401 is the recognition mode. connect. Next, at 402, the main CPU 101 and the microprocessor 110 detect whether the other switches of the camera are turned on, and at 403, detect whether the voice input button 11 is pressed and the SW-NOISE is turned on. I do.

If it is OFF, it returns to 402; if it is ON, it returns 4
Proceed to 04 to start noise detection processing. This 404
After that, the contents of the flowchart up to 413 are SW-VOICE at 408
Is the same as the registration mode 308 to 317 except that the sequence returns to 402 when is OFF, and performs a series of speech recognition processing such as noise spectrum analysis processing, noise removal processing, and speech feature extraction processing. The description is omitted because it is executed. However, the operation of the voice recognition device is the same, but after pressing the second stroke of the voice input button 11 at 408, the photographer does not speak the mode name to be registered, but the photographer is registered in advance. One of the vocabulary words (mode name to be selected) is uttered.

In the next step 414, a matching process is performed between the voice pattern to be recognized, which is obtained by extracting the characteristic amount of the removed voice spectrum pattern, and the reference voice pattern stored in the RAM 114 of the reference voice pattern storage unit 206. Do. The matching calculation is based on the DP (Dynamic Programming) matching method and the HM
This is performed as a distance calculation between the reference voice pattern vector and the voice pattern vector according to a predetermined voice recognition algorithm such as the M (Hidden Markov) method. Next, proceeding to 415, the smallest one of the distances from the reference voice patterns registered by the determination unit 207 is determined as a recognized word,
Output as a recognition result.

Next, the process proceeds to 416, where the reliability of the speech recognition degree is determined. That is, it is determined whether or not the distance between the input voice pattern and the recognized reference voice pattern is smaller than a predetermined reference value. If it is larger, it is determined that there is no recognition reliability, the process proceeds to 417, and a recommendation display is performed from the speaker 14 as "again" so as to perform the input operation again.

If the reference voice pattern is too far away from the reference voice pattern, or if the reliability cannot be obtained no matter how many times, it is possible to make a voice recommendation to "re-register". . If the distance is small, the recognition reliability is judged to be sufficient, and the process proceeds to 418, where the microprocessor 1
10 sends the recognition result to the main CPU 101,
01 switches the camera settings to the shooting mode corresponding to the recognition result, and displays the mode display corresponding to the recognition result on the monitor LCD1.
Display at 09. At the same time, the process proceeds to 419, where the microprocessor 110 generates a standard easy-to-understand sound stored in the ROM 115 in advance corresponding to the shooting mode, and notifies the photographer of the shooting mode.

Thus, a series of voice input operations is completed, and the photographer can shoot in the shooting mode changed by voice.

In the present embodiment, the audio input switch 11 is constituted by a two-click click type tact switch, but the present invention is not limited to this, and the audio input switch 11 can be turned on with a time delay.
A two-contact switch may be used.For example, a so-called electrostatic switch that arranges an electrode with an appropriate gap on the key top and places a contact on the key top with a change in capacitance when the SW-NOISE is touched with a hand. May be used. In this case, the configuration is more complicated than the tact switch, but there is an advantage that environmental noise can be acquired without the photographer being conscious.

As described above, according to the first embodiment of the present invention, the environmental noise is acquired by using the time difference between the first stroke and the second stroke in which the voice input switch is pressed. Only immiscible noise components can be accurately sampled.

Since the environmental noise is collected immediately before the actual input of the voice by the operator, it is possible to accurately collect the environmental noise generated at the time of the voice.

Further, environmental noise is collected during a series of operations of pressing the voice input switch when viewed from the photographer, so that the photographer does not need to perform any special operation or setting.
There is an effect that environmental noise can be accurately and unconsciously collected.

(Second Embodiment) FIGS. 9 to 11 show a second embodiment of the present invention. FIG. 9 is a schematic diagram of a single-lens reflex camera corresponding to FIG. 1 of the first embodiment. FIG. 10 shows the first
FIG. 11 is a block diagram showing an electrical configuration corresponding to FIG. 2 of the first embodiment, and FIG. 11 is a flowchart showing an operation in a recognition mode of the first embodiment.

FIG. 9 differs from FIG. 1 in that it has a known eyepiece detection mechanism for detecting that the photographer's eyes are in the finder section, and emits infrared light, which is a component of the mechanism, from the periphery of the finder. Diode section 6 and phototransistor section that receives the reflected light of the infrared light from the photographer
7 is added.

In FIG. 10, the difference from FIG. 2 is that the SW-NOI
The eye detection circuit 121 and the IRE
Eyepiece detection means 1 composed of D122 and phototransistor 123
20 has been added.

The operation of the speech recognition apparatus according to the second embodiment in such a configuration will be described. The difference between FIG. 11 and FIG. 8 is that “noise input switch =
Instead of the determination of “ON”, the determination of step 503 “eye detection” is included. In this 503, the main CPU 1
01 is the infrared light emitting diode part around the finder
The infrared light is emitted from 6, and the reflected light from the photographer of the infrared light is received by the phototransistor unit 7. Then, if reflected light having a predetermined intensity or higher is detected, it is detected that the photographer is in eye contact with the viewfinder. Send to processor 110.

The microprocessor 110 receives this communication, returns to 502 if turned off, proceeds to 504 if turned on, and starts noise detection processing.

Hereinafter, sequences other than 504, ie, 501
To 502, 504 to 519 are 401 to 402, 404 to 41 of the first embodiment.
Since it is the same as 9, it is omitted.

Although the present embodiment has been described in the case of the recognition mode, the present invention is applied to the registration mode or the mode in exactly the same manner.

As described above, according to the second embodiment of the present invention,
Since the environmental noise is acquired by the eye movement of the photographer, the photographer does not need to be conscious of acquiring the environmental noise at all.

Further, since a sufficient time can be taken before the actual voice input, a sufficient time can be taken for the noise processing time, and in the case of the first embodiment, before the photographer presses the voice input button in the first embodiment. There is an effect that environmental noise can be accurately processed even when uttered.

(Third Embodiment) FIG. 12 shows a camera operation sequence in a camera according to a third embodiment of the present invention, and is a flowchart showing a noise detection operation in the camera sequence.

When the power of the camera is turned on (601), the release button 2 is half-pressed in the next step 602 to detect whether the switch SW1 is ON. If the switch SW1 is OFF, this detection is repeated. The automatic focus detection circuit 102 detects the focus of the photographing lens with respect to the object to be photographed, performs the AF operation, and the light metering circuit 104 measures the luminance of the object to be photographed, and sets the exposure value based on the measured light value. To determine.

Next, at 604, the release button 2 is further pressed, and
Detects whether 2 is ON, and if it is ON, proceeds to 605, executes a known camera release sequence by the shutter control circuit 105, the aperture control circuit 106, and the motor control circuit 107, and prepares for the next operation To return. If it is OFF, the process proceeds to 606, where it is detected whether the photometric value has changed by a predetermined value or more.

This is because when the photometric value has changed by a predetermined value or more (for example, when the photometric value differs by 3 to 4 steps or more), the environment where the photographer is placed may have changed. It is determined that the environmental noise may have changed. If it has not changed, the process proceeds to 607, and if it has changed, the process proceeds to 608 to start noise detection.

At 607, it is detected whether or not the amount of defocus of the photographing lens has changed by a predetermined value or more. This is because when the amount of defocus has changed by more than a predetermined value or more (for example, when the focus at a position of several meters suddenly becomes close or infinity) and the environment where the photographer is placed is 606 It is determined that there is a possibility that the environment may have changed as well as that of the environment. If it has changed, proceed to 608, start noise detection,
If not, return to 602. That is, when either the change in subject brightness or the change in focus is large in the automatic focus detection circuit 102 and the photometry circuit 104, the sequence is to perform noise detection again.

Here, the operation of noise detection after 608 is the first operation.
The operation is the same as that described in the description of FIG.
Accordingly, 608 to 611 are the same as 308 to 311 and will not be described. When the noise spectrum is stored as a noise pattern in 611, the process proceeds to 612, and after it is confirmed that noise has been detected a predetermined number of times, the noise detection is terminated. Next, it is detected whether the sound input button 11 is pressed and the sound SW-VOICE is ON, and the OF
If F, the process returns to 602; if it is ON, the process proceeds to 614 to start voice detection, and recognizes the voice uttered by the photographer. Since the recognition processing has already been described, the description thereof is omitted.

As described above, according to the third embodiment of the present invention,
If the change in the brightness of the subject to be shot or the amount of defocus is detected to be quite large, it is determined that the environment has likely changed, and the environmental noise is acquired. It does not need to be aware that there is no additional cost for detection.

Further, since a sufficient time is taken until the actual voice input, there is an effect that a sufficient time is taken for the noise processing time, and the environmental noise can be accurately processed.

However, since environmental noise is not necessarily collected again when environmental noise changes,
The combination with the second embodiment is appropriate.

(Fourth Embodiment) FIG. 13 shows a fourth embodiment of the present invention.
In the flowchart of the operation in the recognition mode of the camera according to the second embodiment, almost the same operations as those in the recognition mode according to the first embodiment are omitted.

More specifically, the fourth embodiment is a further improvement of the operation after the reliability judgment 416 in the flowchart in the voice recognition mode of the first embodiment of FIG. 8 results in NG. . Note that the same step numbers are applied to those performing the same operation.

In FIG. 8, the speech recognition operation proceeds in the same manner as described in the first embodiment. When the word is determined at 415, the process proceeds to 420, where the reliability of the speech recognition degree is determined, and the input is performed. It is determined whether the distance between the recognized voice pattern and the recognized reference voice pattern is smaller than a predetermined reference value.

If it is larger, it is determined that there is no recognition reliability.
Proceed to 1 and if the distance is short, it is determined that the recognition reliability is sufficient, proceed to 418, switch the camera setting to the shooting mode corresponding to the recognition result, and proceed with 419 and store it in advance corresponding to the shooting mode Generates a standard, easy-to-understand sound.

If the number of times that the recognition result is unreliable at 421 = NG is less than the predetermined number N, the process proceeds to 417, and the first
A re-input recommendation display is made in the same manner as in the embodiment of FIG. If the predetermined number N is not satisfied, the process proceeds to 422, the operation of collecting the environmental noise is again performed, the noise analysis is performed,
Proceeding further to 423, noise spectrum estimation is performed. These are the same as 405 and 406 of the first embodiment. Next, the process proceeds to 424, where the noise spectrum is updated with the newly estimated noise spectrum, or a new noise spectrum is created with the noise spectrum used so far.

In this case, weighting is performed by simply averaging, or increasing the weight of the newest noise spectrum, and sequentially decreasing the weight of the old noise spectrum. The noise spectrum is learned by performing weighting to increase the weight, that is, creating a new noise spectrum by a so-called weighted average. In other words, a new noise pattern is created in consideration of the environmental noise up to the previous time, instead of simply replacing it with a new noise pattern.

Next, the process proceeds to 425, where the newly created noise spectrum is stored, and the process returns to 410, where speech recognition is performed again using the new noise pattern. That is, noise is collected again after the speech recognition result, and the photographer performs speech recognition without generating speech again.

As described above, according to the fourth embodiment of the present invention,
If the input speech recognition result is lower than the predetermined reliability, only the environmental noise is captured again and the speech recognition operation is performed again, so that the recognition rate is improved and the recognition result is NG.
However, it is possible to reduce the need for the photographer to utter voice again, and to learn the environmental noise, thereby making it possible to create a noise pattern that is more suited to the environment where the photographer is placed.

Although the operation in the case where the reliability of the recognition result is low has been described in the present embodiment, the same applies to the case where the recognition result is undefined.

Although the embodiment of the present invention has been described as applied to a single-lens reflex camera, various types of imaging devices such as a video camera and an electronic still camera, and a portable small precision device other than a camera. And other devices.

[0093]

As described above, according to the present invention,
By collecting environmental noise at appropriate timing, environmental noise can be accurately collected, and the recognition rate of the voice recognition device can be improved. Also, the voice recognition device itself determines whether or not environmental noise needs to be collected, and if necessary, automatically collects environmental noise. Therefore, the user performs a special operation for collecting environmental noise. This eliminates the necessity, so that the operation of the speech recognition device can be simplified and the recognition rate of the speech recognition device can be improved.

[Brief description of the drawings]

FIG. 1 is a top view, a rear view, and a side view of a single-lens reflex camera according to an embodiment of the present invention.

FIG. 2 is an essential part cross-sectional view showing a configuration of a switch unit of the voice input button in FIG. 1;

FIG. 3 is a block diagram showing an electrical configuration of the camera shown in FIG. 1;

FIG. 4 is a block diagram of a voice recognition process of the camera in FIG. 1;

FIG. 5 is a flowchart showing the operation of the registration mode of the camera of FIG. 1;

FIG. 6 is a flowchart showing a display state in a photometric mode setting of the camera in FIG. 1;

FIG. 7 is an exemplary view for explaining a noise removal operation performed by the speech recognition apparatus in FIG. 1;

FIG. 8 is a flowchart showing an operation of the camera in FIG. 1 in a recognition mode.

FIG. 9 is a top view, a rear view, and a side view of a single-lens reflex camera according to a second embodiment of the present invention.

FIG. 10 is a block diagram showing an electrical configuration of the camera shown in FIG. 9;

FIG. 11 is a flowchart showing an operation of the camera in FIG. 9 in a recognition mode.

FIG. 12 is a flowchart illustrating a noise removal operation in a camera operation sequence in a single-lens reflex camera according to a third embodiment of the present invention.

FIG. 13 is a flowchart illustrating an operation in a recognition mode in the camera according to the fourth embodiment of the present invention.

[Explanation of symbols]

 5 Metering Mode Setting Button 11 Voice Input Button 13 Voice Mode Switch 14 Speaker 15 Microphone 101 Main CPU 102 Automatic Focus Detection Circuit 104 Photometry Circuit 110 Microprocessor 120 Eyepiece Detection Unit 201 Spectrum Analysis Unit 202 Noise Removal Unit 204 Feature Extraction Unit 205 Recognition (Collation) unit 206 judgment unit

Claims (8)

[Claims] 1. A microphone, an operation member for operating the microphone, first switch means that is turned on by a first stroke of the operation member, and is turned on by a second stroke following the first stroke of the operation member. And a signal component input to the microphone when the first switch is turned on, from a signal component input to the microphone when the second switch is turned on. A voice recognition device comprising: a calculation unit for subtraction; and a voice recognition unit that performs a voice recognition operation using an output of the calculation unit. 2. A microphone, first switch means for operating the microphone when a user's eye is at a predetermined position, and the microphone when the user operates a predetermined operation member. A second switch for operating; and a signal component input to the microphone when the first switch is turned on from a signal component input to the microphone when the second switch is turned on. And a voice recognition unit that executes a voice recognition operation using an output of the calculation unit. 3. A microphone, first switch means for operating the microphone when a body part of the user is at a predetermined position, and when the user operates a predetermined operation member, the first switch means operates the microphone. A second switch for operating the microphone; and a signal component input to the microphone when the second switch is turned on is input to the microphone when the first switch is turned on. A voice recognition device comprising: a calculation unit for subtracting a signal component; and a voice recognition unit that performs a voice recognition operation using an output of the calculation unit. 4. A microphone; first switch means for operating the microphone; second switch means for operating the microphone; and input to the microphone when the second switch means is turned on. Calculating means for subtracting a signal component input to the microphone when the first switch means is turned on from a signal component; and speech recognition means for executing a speech recognition operation using an output of the calculating means. Characteristic speech recognition device. 5. A voice recognition device for detecting an environmental noise signal and performing a voice recognition operation using a signal obtained by subtracting the environmental noise signal from an input voice signal, wherein a space in which the voice recognition device is used is imaged. Imaging means, when an image taken by the imaging means changes,
A speech recognition device comprising: an environmental noise detection unit that detects an environmental noise signal. 6. A speech recognition apparatus for detecting an environmental noise signal and performing a speech recognition operation using a signal obtained by subtracting the environmental noise signal from an input speech signal, wherein a space in which the speech recognition apparatus is used is imaged. A speech recognition apparatus comprising: an imaging unit; and an environmental noise detection unit that detects an environmental noise signal when the luminance of an image captured by the imaging unit changes by a predetermined value or more. 7. A voice recognition device that detects an environmental noise signal and performs a voice recognition operation using a signal obtained by subtracting the environmental noise signal from an input voice signal, wherein an image of a space where the voice recognition device is used is taken. A speech recognition apparatus comprising: an imaging unit; and an environmental noise detection unit that detects an environmental noise signal when a defocus amount of an image captured by the imaging unit changes by a predetermined value or more. 8. A speech recognition apparatus for performing a speech recognition operation by detecting an environmental noise signal and comparing a signal obtained by subtracting the environmental noise signal from an input audio signal with a reference audio pattern, wherein the signal and the reference A reliability judging unit for judging the reliability of the speech recognition degree based on a difference from the pattern; and detecting an environmental noise signal when the reliability judgment unit judges that the reliability of the speech recognition degree is low. A speech recognition device comprising: an environmental noise detection unit.