JP3164346B2 - Voice Recognition Interactive Doll Toy - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice-recognition interactive doll toy, and more particularly, to a voice-recognition dialogue in which a voice-recognition system is provided in a doll toy, and an interesting dialogue with a user can be made by voice expression. The present invention relates to a model doll toy and a control method thereof.

[0002]

2. Description of the Related Art In general, children tend to learn life education through interesting play or toys, and the close contact with the toys is to perform imitation learning that leads to the real world. Such imitation learning is mostly performed through dolls. Therefore, the child makes his or her own imitation learning scenario, and in accordance with the scenario, induces an appropriate response to the doll. You will be immersed in.

[0003] As described above, education using toys has been carried on as a close contact with children since ancient times. In recent years, research on voiced dolls that can be expected to have such educational effects has become active, and attempts have been made constantly to produce more advanced dolls.

Most of such doll toys according to the prior art have a touch sensor provided at a certain position of the doll. When a child activates this touch sensor, a voice expression of a simple sentence recorded on a magnetic recording medium (magnetic tape) or a semiconductor recording medium (IC memory), for example, "Today, I am Mickey. Who are you? "What are you doing?" And so on. In addition, the user performs only a few types of stylized movements, for example, simple movements such as raising the arm and moving the head, to satisfy the temporary curiosity.

[0005]

Therefore, such a conventional doll toy is a spontaneous doll which speaks a simple sentence, and a simple sentence without a scenario is recorded by the operation of a touch sensor. Can trigger a temporary curiosity. However, the child gets tired immediately, and the period of playing with such a doll toy is shortened, so that the educational effect is low.

[0006] In addition, the audio text traditional doll toy is speaking, not an interactive scenario, Ri <br/> Oh in the enumeration of discrete sentences, for poor in reality, also decreases gradually the educational effect There is a problem.

The present invention has been made to solve such a problem. That is, an object of the present invention is to provide a voice recognition interactive doll toy which recognizes a voice of a user such as a child and can continuously perform a dialog according to at least one scenario according to a child's thinking method and behavior style. An object of the present invention is to provide a control method.

[0008] A first object of the present invention is to basically enable voice output according to a topic, create and record a behavior pattern that a child can take in a scenario, and respond to an arbitrarily set situation. To enable two-way conversation with a doll. A second object of the present invention is in situations where children interact,
In order to lead to various scenarios, after compressing the voice with the voice compression software, record it in ROM (ROM) and take it out promptly when necessary. It is an object of the present invention to provide a device that enables immediate dialogue and prompt audio output. A third object of the present invention is to use a speaker-independent voice recognition technique to learn the voices of the multiple people so that a reasonable reaction occurs in order to understand the voices input from the unspecified multiple people. Sometimes. A fourth object of the present invention is to make it possible to discriminate between a surrounding noise and a child's voice by software that can appropriately process the noise by preparing a noise when touching and stroking the doll. A fifth object of the present invention is to provide four contact switches so that when the doll is in a specific posture or when the child contacts a predetermined part of the doll, that is, when the child and the doll are in contact with each other, an appropriate sound is output. there to inspire an interest through the reaction. A sixth object of the present invention is to allow a system to understand an input audio signal,
By properly interpreting this, the hardware is implemented so that an appropriate reaction can be performed in real time, and realistic contents (scenarios) that respond to humans are extracted from a pre-recorded database and output. Is to get it.

Accordingly, the present invention is to realize advanced software and advanced circuit manufacturing technology so as to satisfy such various functions and performances, that is, to provide a doll with a voice decoder, a voice recognition unit, and a system controller. , Dialog managers, and other various auxiliary functions that can elicit interest, mainly satisfy the external and internal requirements of children's doll toys, and enhance the educational effects of language (language education, play education) It is an object of the present invention to provide a voice-recognition interactive doll toy that has speaker independent, artificial intelligence, and interactive performance so that it can be played.

[0010]

According to the present invention, in order to achieve the above-mentioned object, the shape of human and animal is mixed.
A first memory unit (33) in which digital audio signal streams of a large number of sentences are compressed at a predetermined compression rate in the formed doll body, and a user voice input from outside. A speech recognition interactive doll toy comprising a second memory unit (35) provided with an operation area for recognizing a signal, wherein the second memory unit (3)
5) recorded in, and outputs the converted electrical voice signal to the voice of the user of the at least one sentence, voice input and output unit to aurally tell the decompressed audio signal to the user (3
7) and a frame output from the audio input / output unit (37).
Temporarily record the digital audio signal of the user
Circular buffer (51), and the circular buffer
The digital audio signal recorded in the
Voice recognition of compressed data recorded in one memory unit (33)
Viterbi Argo divided into words for speech recognition according to the
A voice recognition unit (5) that recognizes the user's voice in rhythm
3) and the speech recognized by the speech recognition unit (53).
At least one so that the content corresponds to the given scenario
For selecting a response sentence in the first memory unit (33).
Alog manager (55) and the dialog manager
Sound of the first memory unit (33) selected by the keyer (55)
Voice decoder for expanding and restoring voice compressed data (57)
And the audio decoder (57) and the audio input / output unit (37)
Analog audio signal and digital audio signal provided between
A / D / D / A converter (4
7), the second memory unit (35) and the list control
And between the first memory unit (33).
A memory controller for transferring data to the second memory unit (35);
And a controller (63) .

[0011] In the configuration of the present invention, voice output according to the topic is enabled, action patterns that a child may perform are created and recorded in a scenario, and bidirectional communication with the doll can be performed according to an arbitrarily set situation. Dialogue becomes possible. For example, in order to lead to various scenarios in a situation in which the child interacts with the child, the voice is compressed by voice compression software, and then the voice is recorded in the first memory unit (33). Even in one topic, questions and answers can be made immediately according to the available situations. In addition, this speech recognition interactive doll toy implements hardware so that the system can understand the input voice signal, interpret it appropriately, and perform an appropriate reaction in real time, and it is recorded in advance. From the database, realistic contents (scenarios) to which humans react can be extracted and output.

[0012]

[0013]

[0014]

[0015] The first memory section (33) and the voice recognition section (53) are connected between the dialog manager (55) and the first memory section (33) . The voice compression data and the voice recognition constant of the compressed data are taken out from the memory unit (33) , and the second memory unit (35)
List controller for transferring the voice recognition data to the memory
(61) is preferably provided.

The voice recognition section (53) removes predetermined noise from the digital voice signal in frame units recorded in the circular buffer (51) by using a voice recognition constant of the first memory section (33). A speech recognition calculation unit (71) for calculating an eigenvalue for one character voice as a feature vector, and a zero-crossing rate (73) for detecting zero points from the sampling value of the digital speech signal.
When the zero crocin order to improve the reliability of the 0-point detection of the Great (73), an energy calculation unit for calculating an energy for the zero point (75), wherein the energy calculating unit and the zero crocin Great (73)
(75) a unit voice detection unit (77) for detecting end point data of any one word in a continuous digital voice signal based on the output signal of (75) ; and feature vector data of the voice recognition calculation unit (71). And a preprocessor (79) for dividing the words into speech recognition words one by one based on the end point data of the unit speech detector (77) , and a first processor corresponding to the words divided by the preprocessor (79) . A second memory section (35) for providing an area in which the compressed voice data of the memory section (33) is extracted by the list controller (61) and operated by the Viterbi algorithm.

With such a configuration, in order to understand voices input from an unspecified number of people, the voices of the plurality of people are learned by a speaker-independent voice recognition technique so that a rational reaction occurs. Can be

[0018] The are embedded into a plurality of regions of the doll, to inform the contact of the user to the audio decoder (57), the doll's back, nose, mouth and contact <br/> provided ass touch It is preferable to further include a switch (T1, T2, T3, T4) .

When the user makes contact with the contact switches (T1, T2, T3, T4) provided on the mouth, nose, back and buttocks, an appropriate sound corresponding to the contact is transmitted to the dialog manager (55) and the first switch. It is preferable that the audio data is extracted from the memory unit (33) , decompressed and restored to the actual audio by the audio decoder (57) , and then audibly heard by the user through the audio input / output unit (37) .

[0020] Then, the voice output unit (37), the first micro <br/> phone for outputting voice and external noise of the user on said converted into an electrical signal circular buffer (51) ( 39) , a second microphone (41) for converting the external noise into an electric signal and outputting the electric signal to the circular buffer (51) , and expanded and restored by the audio decoder (57) . Power amplification of audio signal, speaker
(43) a power amplifying section (45) for hearing the user audibly.

The conversion between the said circular buffer (51) first and second microphones (39, 41), the output signal of the <br/> first and second microphones (39, 41) into a digital An A / D / D / A converter (47) for converting the digital audio signal expanded and restored by the audio decoder (57) into an analog signal between the audio decoder (57) and the power amplifier (45 ); preferably are al provided.

Thus, the four contact switches (T1, T1,
T2, T3, T4) , whether the doll is in a specific posture,
Idaku when the child comes into contact with a predetermined portion of the doll, i.e. when there is contact between the child and the doll, the interest through the appropriate voice reaction
You can make it.

The A / D / D / A converter (4)
7) and the power amplifying unit (45) according to a command for adjusting the volume of the user (“louder sound” and “lower sound” ). It is preferable that a volume control unit (49) for adjusting the output intensity is provided.

[0024]

[0025]

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A speech recognition interactive doll toy of the present invention will be described below in detail with reference to the accompanying drawings. FIG.
FIG. 2 is a front view showing the doll toy of the present invention, FIG. 2 is a side view showing the doll toy of the present invention, FIG. 3 is a system block diagram showing the voice recognition interactive doll toy of the present invention, and FIG. FIG. 5 is a block diagram of an ASIC-based voice recognition unit of the present invention, and FIG. 6 is a flowchart showing the operation of the voice recognition interactive doll toy of the present invention.

As shown in FIGS. 1 and 2, the doll toy of the present invention has the shape of a sewing doll, is covered with a skin, and the inside of the head and the inside of the body are provided for protecting the circuit. ,
It consists of a structure with a solid frame (not shown).

The illustrated example has a fairy shape which resembles a human as a whole, and has a torso and a chest 1, two hands 2 and 3 having four fingers, and two arms 8, 9. The lower body consists of two legs 4,5, both feet 6,7 with four toes, buttocks and tail 17. Face is mouth 1
0, both ears 11, 12, hair 16, and both eyes 14, 15
It is composed of

As shown in the side view of FIG. 2, the neck 19 connecting the head and the body is made of a flexible material which can be bent to facilitate the wiring between the head circuit and the body. Further, the doll toy of the present invention has a beautiful appearance and is covered with a material capable of protecting the internal circuit, and is formed into a soft and realistic child doll shape.

The doll toy is provided with four contact switches for inducing a reaction of the doll when it comes into contact with the doll. The arrangement position is nose (T1), mouth (T
2) The back (T3) and buttocks (T4). These contact switches T1, T2, T3, T4 are manufactured so that contact can be easily detected. These contact switches T1, T2, T3, and T4 have extremely high sensitivity, are provided inside the skin of the sewing doll, and induce a voice response when touched.
C; custom-made semiconductor-microprocessor) is directly input with a high active signal. In particular, the contact switch T4 on the buttocks has a function of detecting whether the doll is standing or sitting and operating an appropriate response.

[0031] For example, the reaction caused by the contact switch T4 of the buttocks, when you are taking to have attitude to the prone position is, "Do not rest together", when in a standing posture,
"I want to play", touch the contact switch T2 on the mouth, "Very delicious", release my hand from the contact switch T2, "I'm stomach stomach", touch the contact switch T3 on the back, "Who is it?" Or touching the contact switch T1 on the nose, a sound "Tickling, er" is output.

The doll toy system of the present invention is shown in FIG.
As shown in the figure , a circular buffer 51 and a voice recognition unit
53, an audio decoder 57, and an A / D / D / A converter
A voice processing control unit (ASIC-application specific IC) 30 including a data controller 47, a memory controller 63 , a first memory unit (ROM) 33, a second memory unit 35, and a voice input / output unit 37. I have. The first memory unit 33 stores compressed audio data obtained by compressing a digital audio signal stream of a large number of sentences at a predetermined compression ratio. The second memory unit 35 stores an externally input audio signal of the child. An operation area for recognition is provided, and the voice processing control unit 30 is provided.
Analyzes the interactive response corresponding to the child's voice signal using the calculation area of the second memory unit 35, and decompresses and decompresses the voice compressed data corresponding to the response from the first memory unit 33. The voice input / output unit 37 converts at least one sentence of the child's voice into an electrical voice signal and outputs the electrical voice signal to the voice processing control unit 30. The voice processing control unit 30 converts the amplified voice signal into the electrical voice signal. Tell the child aurally.

As shown in FIG. 4, the voice input / output unit 37 converts a voice of the child and noise generated from the skin of the doll into an electric signal and outputs the electric signal to the circular buffer 51. When the circular converts the noise generated from the epidermis of the doll into an electrical signal
A second microphone 41 for outputting to the buffer 51 and a power amplifying unit 45 for power-amplifying the expanded and restored voice signal from the voice decoder 57 and hearing it to the child through the speaker 43 are provided. The circus
Buffer 51 and the first and second microphones 3
Between the 9,41, and converts the output signal of the first and second microphones 39 and 41 to digital, the audio de
Between the coder 57 and the power amplifier 45, the audio de
The coder 57 is provided with an A / D / D / A converter 47 for converting the expanded and restored digital audio signal into analog. Here, the speaker 43 is configured to process the compressed voice recorded in the first memory unit 33 in a predetermined process and to allow the user (child) to hear the processed voice.

On the other hand, the volume control unit 49 for adjusting the output intensity of the power amplifying unit 45 so as to increase the sound actually generated from the speaker 43 is provided by the A / D.
It is connected between the D / A converter 47 and the power amplifier 45. For example, the volume control unit 49
In order for the child to adjust to the desired sound volume, the first
When a command for volume adjustment (for example, “louder sound” and “lower sound”) is input through the A / D / D / A converter 47 through the microphone 39, the power amplifying unit 45 is controlled. And the speaker 43
, So that sound of a volume corresponding to the command is generated. As a result, the magnitude and gain of the power amplifying unit 45 are determined based on the unmute signal of the system controller 59 of the audio processing control unit 30 and the output signal of the volume control unit 49.

The first and second microphones 39 and 41 of the audio input / output unit 37 are provided to have a noise removing function.
9, a signal in which voice and noise are mixed is input,
A pure noise signal is input to the second microphone 41 when the doll comes into contact with the user or is affected by ambient noise. At this time, the audio processing control unit (ASIC) 3
0 correlates a voice and noise signal through the first microphone 39 with a pure noise signal through the second microphone 41 to reduce noise using a correlation (function) between noises of both signals. To remove only the noise component. The first and second microphones 39 and 41 are mounted on both ears 11 and 12 of the doll based on experimental results. In particular, one of the first and second microphones 39 and 41 is small. And a stereo microphone having a high sensitivity and a high directivity in an audio frequency band is provided.

Each contact switch T1, T2, T3
T4 is directly connected to the audio processing control unit 30, as shown in FIG. The circular buffer 51 temporarily stores a digital audio signal of a child input to the audio input / output unit 37, that is, an audio sampling signal digitized in frame units converted by the A / D / D / A converter 47. The voice recognition unit 53 records the digital voice signal recorded in the circular buffer 51 into voice recognition words based on the voice recognition constant of the compressed data stored in the first memory unit 33, and to recognize the meaning of the child's voice in the algorithm, the dialog
The manager 55 selects any one of a number of scenarios in which the content of the voice recognized by the voice recognition unit 53 is expanded, and compresses at least one sentence of the compressed voice data so as to correspond to the selected scenario. The first memory unit 33
And Eject from the audio decoder 57, the dialog <br/> stretching and restored to the audio compression data retrieved from the grayed manager 55 outputs to the audio output unit 37. A system controller 59 for outputting appropriate control signals to the first memory unit 33, the second memory unit 35, the volume control unit 49, the A / D / D / A converter 47, and the power amplification unit 45 is provided. .

When the child makes contact with the contact switches T1, T2, T3, and T4 installed on the nose, mouth, back, and buttocks of the doll, the audio processing control unit 30 generates appropriate compressed audio data corresponding thereto. The dialog manager 55
Then, after being extracted from the first memory unit 33, converted by the audio decoder 57 and expanded and restored to an actual audio, control is performed so that the child can be audibly heard through the speaker 43 of the audio input / output unit 37.

In the first memory 33, speech and music of many sentences according to a scenario, a large number of speech data, speech recognition constants, and restoration data for speech decoding are compressed and recorded. The element used has a large capacity of 4 Mbytes or more, and can record data in units of 1 word (16 bits), so that a total of 2 Mwords can be recorded. The recorded contents are shown in Table 1 below.

[0039]

[Table 1] Breakdown of information recorded inside the first memory unit

The second memory unit 35 is an element for recording a processing program for processing a child's voice and a voice of a response sentence and performing internal data signal processing. The second memory unit 35 is used for a block list. It has an area and a use area for preprocessing of voice recognition, and has a predetermined data storage capacity. Here, the list controller 61 not only function to retrieve the data of the second memory unit 35, and outputs it to the audio decoder 57 also takes out the compressed audio data of the first memory unit 33.

Here, between the second memory unit 35 and the list controller 61 of the audio processing control unit 30,
For data transmission between each other, a memory controller 63 is provided. In particular, the memory controller 63 is configured to read the compressed data and output it to the second memory unit 35.

On the other hand, the power supply unit 65 maintains an arbitrary voltage within a voltage fluctuation range of 3 to 24 V at a constant voltage of 3.3 V, and basically includes three dry batteries (4.5 V) connected in series.
, But other power sources can be used. Elements required for other operations include:
The second memory unit is indispensable for operating the above components such as the clock generator 67 of 24.546 MHz for 35 clocks and the timer 69 of 32.768 kHz, and the description thereof will be omitted.

As shown in FIG. 5, the voice recognition section 53 removes a predetermined noise from the digital voice signal of the frame unit recorded in the circular buffer 51 by using the voice recognition constant of the first memory section 33. And a speech recognition calculation unit 71 for calculating an eigenvalue for one character as feature vector data, and a zero-crossing rate 73 for detecting 0 points from the sampling value of the digital speech signal.
And an energy calculator 75 for calculating the energy for the zero point in order to improve the reliability of detecting the zero point at the zero crossing rate 73, based on the zero crossing rate 73 and the output signal of the energy calculator 75. Which one of the continuous digital audio signals
A unit speech detection unit 77 for detecting end point data of a word; and a preprocessing for dividing words into words for speech recognition one by one based on feature vector data of the speech recognition calculation unit 71 and end point data of the unit speech detection unit 77 And a second memory unit 35 for extracting the compressed voice data of the first memory unit 33 corresponding to the words divided by the pre-processor 79 by the list controller 61 and calculating by the Viterbi algorithm. Here, a pre-emphasis 81 that amplifies the frequency in order to process the digital audio signal of the circular buffer 51 more quickly is provided between the audio recognition calculator 71 and the circular buffer 51.

More specifically, the voice recognition unit 5
The calculation flow and the module configuration of No. 3 consist of two module groups, and are composed of many server modules in which the Viterbi algorithm and the unit voice detection algorithm are integrated by ASIC semiconductors.

First, the Viterbi algorithm starts at 4 years old
H, so that it can be used in children's toys of the age of 0
It is composed of one chip using a Viterbi algorithm using MM (Hidden Markov Model). Also, the external second memory unit 35 is designed to process a large amount of variable data generated in the process of executing the Viterbi algorithm.
(16 Mbytes), and uses a block list structure capable of accommodating about 16 Mbytes.
It is configured to operate in the region. The HMM learning method is to improve the reliability even if the user is different, that is, to perform speaker-independent recognition, and to recognize phoneme units.

With reference to FIGS. 3, 4 and 5, the operation of each element described above will be briefly described as follows. First, two microphones 39 and 41 is converted into an electric signal by receiving the audio signal and sends the analog sound signal conversion apparatus A / D · D / A converter (Codec) 4 7 described above.
At this time, the two input audio signals are transmitted to the audio processing controller 30 in a form independent of each other to remove noise, and are correlated. The voice processing control unit 30
Then, unless there is a special situation, a control signal (data input preparation signal) is sent to the A / D / D / A converter 47,
Informs that it is now receiving ready, for between complement in A / D · D / A converter 47 (Interpolation) x25
The value of 6FS is 2.048 MHz, and the synchronization frequency (SYNC Frequency) is 8 kHz.
In 3 it is applied at a sampling rate that improves speech recognition. In particular, the 8 kHz sampling rate is an important processing criterion for a recognition algorithm in the voice recognition unit 53 of the voice processing control unit 30. On the other hand, the input audio signal is converted by an A / D / D / A converter 4.
The signal is A / D-converted at 7 and sent to the audio processing control unit 30, input as independent data through the first and second microphones 39 and 41, and the noise due to the correlation operation is filtered.

As described above, the digital audio sampling signal from which noise has been removed is temporarily recorded in a frame unit in the circular buffer 51, and then the pre-emphasis 81 and the audio recognition calculation unit 71 output the digital audio sampling signal from each user. An eigenvalue for speech is calculated as a feature vector, and a zero crossing rate 73 is used to detect the end point of each word.
And the energy calculation unit 75 and the unit sound detection unit 77 almost simultaneously.
In step 9, the words are classified into words for speech recognition. Then, when the compressed data of the first memory unit 33 corresponding to the speech recognition word from the preprocessor 79 is extracted from the list controller 61, the data and the Viterbi algorithm are transferred to the second memory unit 35, and the recognition is performed. And perform an arithmetic operation for the analysis.

More specifically, the operation actually comprises the steps of an audio signal sampled at 8 kbps → preprocessing (voice feature detection) → voice detection → voice recognition. The pre-processing is performed on the spectrum results that have been subjected to calculation steps such as Power, Hamming Window, and Pre-Emphasis, and then subjected to Real FFT.
Calculate the cepstrum of the scale. Separately, the speech is calculated by the zero crossing rate 73 and the energy calculation unit 75, and the start point and the end point of the speech are detected. Based on these two voice detection results, the start and end of voice recognition or the presence or absence of reset is determined, and voice is finally recognized using the Viterbi algorithm for the mel-scale cepstrum coefficient sequence and the HMM. Of course, constants necessary for performing such a large number of calculations are recorded in the first memory unit 33, and are taken out and used each time they are needed. Further, the second memory unit 35 is used for the operation of calculating and taking out a necessary value, and the list controller 61 is used for the enormous amount of data calculation. Here, for speech recognition and compression, the end point detection of the speech is performed by the unit speech detection unit 77 used to increase the recognition rate and the compression rate.

Incidentally, the zero crossing rate 73 and the energy calculating section 75 show high efficiency and accuracy in a laboratory or a relatively quiet room, but there is a fundamental problem in the detection of a speech endpoint which responds to even a slight noise. Must be operated with the mel cepstrum.

That is, when a sampling signal obtained by mixing voice, noise, and silence is input to a unit voice detection unit to obtain an energy detection unit, a zero crossing rate, and a mel-scale cepstrum, a voice (noise mixing) portion is output. . Thus, the results from the two modules are sent to the preprocessor 79 to recognize the audio signal.

As described above, when the voice of the user, that is, the child, is recognized by the voice recognition unit 53, the dialog manager 55 divides the recognized voice into one of the scenarios divided into a number of patterns. The compressed data of the response voice according to the selected one scenario is extracted from the list controller 61 and the first memory unit 33 and transmitted to the voice decoder 57.

Next, the audio decoder 57 expands the compressed data in the first memory unit 33 by a predetermined decoding process to restore it to a digital audio signal, and makes the audio input / output unit 37 listen to the child as a speaker. . At this time,
Since the A / D / D / A converter 47 is provided between the audio decoder 57 and the audio input / output unit 37, the decoded digital audio signal is changed to analog and generated as actual audio. Is done.

Here, when a command to "increase the sound" is input from the child's voice through the voice input / output unit 37 to adjust the volume, the command is A / D
-It is input to the voice processing control unit 30 via the D / A converter 47 and recognized. Then, according to the volume control signal of the audio decoder 57 , the predetermined gain is determined by the volume control unit 49, and the A / D / D / A converter 4
The power amplifier 45 is controlled so that the analog audio signal output to 7 is made larger than a normal amplification gain value so as to be heard by a child.

The speech recognition interactive doll toy according to the present invention as described above is operated by a control method as shown in FIG.
Make. First, the stages to which the doll can react are classified into three stages (battery ON, time signal, contact switch mode) (130, 131, 132). If, that power is supplied, the battery power is replaced, a predetermined greeting of the word, that is, "Hi, I'm Sarah. Who are you do." Word of greeting is output that (130 ), A message recorded by the user, that is, a greeting message of a message that differs depending on the morning, noon, and evening, or a word according to a set scenario is output as a voice (timer mode, message mode).
(131), a voice corresponding to the greeting may be output by the contact switches T1, T2, T3, T4 (132).

According to this flowchart, after the greeting is output by initialization (133), a voice signal is waited for to recognize the child's dialogue voice (134). If the waiting time during which the child does not respond becomes long (137), the time is ended (144), and first, onomatopoeia, song mode, talking mode, and play mode (hospital play, play house, market play) that indicate that the child is waiting. , A tea party play) or output a guidance voice for these modes (145). At this time, the waiting time is about 10
On the order of seconds. If such a process is performed at least three times or more, the save mode mode is set (147). If not, the process waits for the child's voice (146).

Next, as soon as the child makes a desired voice response, a greeting is output, and the child moves to the desired play mode (136, 138) for each play (141). If there is no response or recognition is not possible (unrecognized), the question of the determined method is repeated (143).
As soon as the desired play mode is recognized, a sound indicating the start is output (148), and play is started.
After this step, the user and the doll can proceed in succession according to various patterns (149, 150,
151, 153, 154). Even if recognition is not possible during the dialogue, a decision algorithm is created and calculated so that a person performs a similar action when he or she cannot recognize the other party's words, and then asks again or intentionally It is decided whether to make a possible response on the pattern or to proceed with the play again (153, 157). This depends on the processing method built into the doll. After playing through various patterns,
The user should choose to play again or stop again (155, 156, 158). There are four such play modes, and the number of possible patterns is about 3,000
Leads to seeds.

There are a timer mode and a message recording / playback mode to make the user more interested and emphasize the practicality. All adjustments are made only by the voice recorded inside the doll, and the user can judge the accuracy of the setting by the output voice.

The timer mode uses a three-bit signal from four contact switches and has seven setting modes.
There are eight modes, time adjustment mode. In addition, there is a default setting function in case the power supply is interrupted and the data set by the user disappears. In the setting mode, the setting state is turned on when pressed once, and turned off when pressed twice. If no operation is performed for more than 5 seconds, the camera is automatically turned off with the original state recorded. There are eight modes that can be set: no-action mode, time adjustment mode, wake-up time, breakfast time, lunch time, nap time, dinner time, and bedtime.Each mode determines the user's voice message. It can be set to be output at the specified time. At this time, the time adjustment mode is
It can be set in minutes and, after correction, announces the result by voice. Of course, when it is desired to operate only a part of such functions, the setting mode can be sequentially pressed to turn on / off. Even if the execution is stopped halfway, the value of the existing execution is automatically recorded,
Turned off. When the mode adjustment switch is pressed once, a sound indicating time correction is output, and when the mode adjustment switch is pressed again, a mode for setting the wake-up time is set, and the rest is performed in the same format.

Using the message recording (playback) mode, the user's voice can be recorded for about 4 minutes, and if the recording is made longer, the recording is automatically interrupted. On the other hand, the eight modes can reproduce a desired sound in accordance with seven user setting times. If no adjustment is made within 10 seconds, it will be turned off automatically. Even if the execution is stopped in the middle, the automatically executed value is recorded and turned off.

The audio decoder 57 for converting the compressed audio data into an audio signal has the compressed audio information (144 bits / 240 samples, 16 kHz) recorded in the first memory unit 33.
This module decodes sample data) by a given algorithm, and is composed of a total of 14 submodules, and proceeds in the following order. An initial value is assigned to the second memory unit 35 (set init), first memory unit 3
3 compressed data is read (rd dat), monotone, lsf
After performing LSF processing using int, con gain, stoch c
w 、 adapl cw, adapt2.cw, lsf pc, lp syn, post An audio signal is output through the fit processing module.

[0061]

As described above, the voice-recognition interactive doll toy of the present invention combines a system composed of voice recognition means and voice utterance means with a dialog manager which enables scenario development, for dialogue. In consideration of being a doll for children, a mechanical structure system is put into a sewn doll to induce the desire to play with this doll toy, and the educational effect of language There are excellent effects such as enhancing

[Brief description of the drawings]

FIG. 1 is a front view showing a doll toy of the present invention.

FIG. 2 is a side view showing the doll toy of the present invention.

FIG. 3 is a system block diagram showing a voice recognition interactive doll toy of the present invention.

FIG. 4 is a flowchart according to the processing order of FIG. 3;

FIG. 5 is a block diagram of an ASIC-based voice recognition unit of the present invention.

FIG. 6 is a flowchart showing the operation of the voice recognition interactive doll toy of the present invention.

[Explanation of symbols]

 Reference Signs List 30 audio processing control unit 33 first memory unit 35 second memory unit 37 audio input / output unit 39 first microphone 41 second microphone 43 speaker 45 power amplification unit 47 A / D / D / A converter 49 volume control unit 51 circular buffer 53 voice recognition unit 55 dialog manager 57 voice decoder 59 system controller 61 list controller 63 memory controller 65 power supply unit 67 clock generation unit 69 timer 71 voice recognition calculation unit 73 zero crossing rate 75 energy calculation unit 77 unit voice detection unit 79 preprocessing Container 81 Pre-emphasis T1, T2, T3, T4 Contact switch

Continuing on the front page (72) Inventor Park Young-jeong 28, Baek-ri-ri, Hwasu-si, Gyeonggi-do, Republic of Korea (72) Inventor Kim Unjah Apartment 208-1201, Taekwha-dong, Mapo-gu, Seoul, South Korea 208-201 (72) Inventor Kuon Sapbon 805-402, Fujimura-dong, Gangseo-gu, Seoul, Republic of Korea (72) Inventor E Che-kyon 786-22, Hansa 1-dong, West-gu, Gwangju, Jeollanam-do, Republic of Korea (72) Inventor Chi Kyungchea, Masan, Gyeongsangnam-do, Republic of Korea 509-11, Kaiyuan 2-dong, Ichimaki-en-ku, Japan (72) Inventor Pan Tae-sik 310-1 Kinsuk-dong, Namyangsu-si, Gyeonggi-do, Republic of Korea (72) Inventor Han Chu-yeon Heppima-ul, Koshin-dong, Deok-yang, Goyang-si, Gyeonggi-do, Korea No. 902, 1904 (56) References JP-A-10-179941 (JP, A) JP-A-9-326856 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) A63H 1 / 00-37/00 G10K 15/04 G10L 15/00 G10L 15/28

Claims (8)

(57) [Claims] (1) A mixture of human and animal forms is formed.
A first memory unit (33) in which compressed digital audio signal streams of a large number of sentences are recorded at a predetermined compression rate, and a user audio signal input from the outside. A speech recognition interactive doll toy comprising a second memory unit (35) provided with a calculation area for recognizing the at least one , wherein at least one of the doll toys recorded in the second memory unit (35) is provided.
Converts the speech sentence of the user into an electrical audio signal output, the audio output unit to aurally tell the decompressed audio signal to the user and (37), the voice input and output unit (37 ) Output frame unit
To temporarily record the digital audio signal of the user
And a digital buffer recorded in the circular buffer (51).
The audio signal is transmitted to the first memory section (33).
The words for speech recognition are separated by the speech recognition constants of the compressed data.
Recognize the user's voice using the Viterbi algorithm
A voice recognition unit (53), and the contents of the voice recognized by the voice recognition unit (53) are predetermined.
At least one response statement to support the scenario
Dialog box for selecting a chapter in the first memory section (33)
Manager (55), and the second selected by the dialog manager (55).
Decompress and decompress audio compressed data in one memory unit (33)
Audio decoder (57), and between the audio decoder (57) and the audio input / output unit (37).
An analog audio signal and a digital audio signal
A / D / D / A converter for converting one to the other (47)
And the second memory unit (35) and the list controller (6).
1) The data of the first memory unit (33) provided between
To transfer to the second memory section (35)
Speech recognition interactive doll toy, which comprises a chromatography La (63), the. And a dialog manager (5 ) between the voice recognition unit (53) and the first memory unit (33).
5) and the first memory unit (33) .
A list controller (6 ) that extracts voice compression data and voice recognition constants of the compressed data from the memory unit (33) and transfers the voice recognition data to the second memory unit (35).
The voice recognition interactive doll toy according to claim 1 , wherein 1) is provided. 3. The voice recognition section (53) converts the digital voice signal in frame units recorded in the circular buffer (51) into the first memory section (3).
The predetermined noise is removed by the voice recognition constant of 3) ,
A speech recognition calculation unit (71) for calculating an eigenvalue for one character voice as a feature vector; a zero crossing rate (73) for detecting 0 points from the sampling value of the digital voice signal; and a zero crossing rate (73) . In order to improve the reliability of detecting the zero point, an energy calculating unit (75) for calculating the energy for the zero point, based on the zero crossing rate (73) and the output signal of the energy calculating unit (75) , A unit speech detection unit (77) for detecting end point data of any one word in the continuous digital speech signal; a feature vector data of the speech recognition calculation unit (71) ; and a unit speech detection unit (77) . of Category preprocessor for dividing words for speech recognition word by word based on the end point data and (79), with said preprocessor (79) The first corresponding to the word was
A second memory unit (3) which provides an area in which the compressed voice data of the memory unit (33) is extracted by the list controller (61) and operated by the Viterbi algorithm
5. The voice-recognition interactive doll toy according to claim 1 , comprising: 4. is embedded into a plurality of regions of the doll body, to inform the said contact of the user audio decoder (57), the doll's back, nose, contact provided in the mouth and Ass <br / The voice recognition interactive doll toy of claim 1, further comprising a touch switch (T1, T2, T3, T4) . 5. When the user makes contact with contact switches (T1, T2, T3, T4) provided on the mouth, nose, back, and buttocks, an appropriate sound corresponding to the contact switches is transmitted to the dialog manager (55) and the dialogue manager. The audio data is extracted from the first memory unit (33) , decompressed and restored to the actual audio by the audio decoder (57) , and then audibly heard by the user through the audio input / output unit (37). Claim 4
Voice recognition interactive doll toy. Wherein said voice input portion (37), the first microphone <br/> Rofon for outputting sound and external noise of the user on said converted into an electrical signal circular buffer (51) ( and 39), said circular converts the external noise into an electrical signal
The second microphone (4 ) outputting to the buffer (51)
1) and a power amplifying unit (45) for power-amplifying the audio signal expanded and restored by the audio decoder (57) and hearing it to the user audibly through a speaker (43). The speech recognition interactive doll toy according to claim 1, characterized in that: 7. A circuit between the circular buffer (51) and the first and second microphones (39, 41) .
The first and second output signals of the microphone (39, 41) into digital, said power amplifier and the audio decoder (57) (45) between the speech decoder (57)
In stretching and restored digital audio signal is converted into an analog A / D · D / A converter (47) is provided we are characterized by being claim 6 of the speech recognition interactive doll toy. 8. The A / D / D / A converter (47)
And the power amplifying unit (45) , the power amplifying unit according to a command for adjusting a volume of a user.
Volume control unit (4 ) for adjusting the output intensity of (45)
The voice recognition interactive doll toy according to claim 7 , wherein 9) is provided.