JP3468389B2 - Speech recognition dialogue apparatus and speech recognition dialogue processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enhance speech recognition ratio producing by difference in age and sex and make conversation suitable for sex and ages possible by making a ROM part for memorizing memory contents previously set in order to conduct speech recognition and response output a cartridge. SOLUTION: In a device which analyzes a voice inputted and generates voice characteristic data, compares it with standard voice characteristic data previously registered, outputs word detected data, understands the meaning of inputted voices, decides response contents corresponding to the inputted voices, and inputs them, a memory means which memorizes memory contents previously set in order to conduct speech recognition and memory contents previously set in order to conduct response output to the speech is arranged on a detachable cartridge 20 side to a device main body 1, and connected to a speech recognition response processing part 10 installed on the device main body 1 side by mounting the cartridge 20 on the device main body 1 side, and response contents to input speech are outputted based on data memorized in the cartridge 20.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition dialogue apparatus and a speech recognition dialogue processing method for recognizing speech and performing a response or a specific operation corresponding to the recognition result.

[0002]

2. Description of the Related Art In this type of voice recognition device, there are a specific speaker voice recognition device capable of recognizing a voice of a specific speaker only and an unspecified speaker voice recognition device capable of recognizing a voice of an unspecified speaker. .

The specific speaker voice recognition device registers a standard voice signal pattern of a specific speaker by inputting words that can be recognized by a specific speaker word by word according to a predetermined procedure. After the registration is completed, when the specific speaker speaks the registered word, the speech recognition is performed by comparing the characteristic pattern obtained by analyzing the input voice with the registered characteristic pattern. A voice recognition toy is an example of this type of voice recognition dialogue device. For example, about 10 words such as “Good morning”, “Good night”, and “Hello” are used as a plurality of command words that serve as voice commands, and the child who uses the toy has registered in advance, and the speaker can say “Good morning”. "
Then, the voice signal is compared with the voice signal of the registered "Good morning", and when both voice signals match, a predetermined electric signal for the voice instruction is output, and based on this, the toy is specified. It is what makes the operation.

Such a specific speaker voice recognition device recognizes only a specific speaker or a voice having a voice pattern close to that of the specific speaker, and it is necessary to register all the words to be recognized one by one as an initial setting. The trouble was extremely troublesome.

On the other hand, the unspecified speaker voice recognition device uses the voices uttered by a large number of speakers (for example, about 200 people) to previously generate the standard voice feature data of the recognition target word as described above. It is created and stored (registered), and the speech uttered by an unspecified speaker can be recognized for these pre-registered recognizable words.

[0006]

Although the unspecified speech recognition device capable of recognizing the speech of the unspecified speaker surely has a relatively high recognition rate for the standard speech, it is almost the same. A high recognition rate is not always obtained for all the voices. For example, voice characteristics vary greatly depending on age and gender, such as infant voice, adult voice, female voice, male voice, etc. Even if a very high recognition rate is obtained for adult questions, There is also the problem that little attention is given to infants' questions.

When this type of voice recognition device is applied to a toy such as a stuffed toy, the contents of the dialogue usually change depending on the age and sex of the child who plays the stuffed toy. For example, the contents of dialogues required for infants and upper grades of elementary school and for boys and girls are generally different.

However, in this type of voice recognition device, the number of registered words that can be recognized is also limited, and the response content to that is generally limited to some extent. . Therefore, this type of toy usually gets tired in a short period of time, and as described above, there is a problem in whether the recognition rate is good or bad due to the characteristics of voices such as age and sex. This is true not only for toys but for all electronic devices that utilize voice recognition.

The present invention has been made to solve these problems, and the ROM portion such as standard voice characteristic data storage means for storing standard voice characteristic data and response content storage means for storing response content is a cartridge type. By selecting the cartridge appropriately and mounting it, it is possible to recognize it according to the characteristics of voice due to differences such as age and sex, improve the recognition rate, and support various ages and sexes. The purpose is to enable conversation.

[0010]

A speech recognition interactive apparatus of the present invention analyzes speech input from a speaker to generate speech feature data, and recognizes this speech feature data and pre-registered recognizable words. output word detection data is compared with the standard voice characteristic data, it receives this word detection data, to understand the meaning of the input voice, the voice recognition interaction device for determining and outputting the response content corresponding thereto, certified a plurality of detachably mounted to said speaker age or apparatus main body having a storage means in association with sex, etc. storing preset data in order to perform a response output to identify speech The cartridge is provided, and at least one selected from the plurality of cartridges is attached to the apparatus main body to connect to the voice recognition response processing unit provided on the apparatus main body side. The dan is for instructing what kind of speech synthesis output should be generated, and conversation content storage means that stores the response content corresponding to the recognizable word registered in advance corresponding to the speaker's age or sex. Response data instruction content storage means for storing the instruction content of the voice recognition response processing section, the voice recognition response processing section, the state detection means, the voice input means for inputting the voice of the speaker, and the voice input means. Voice analysis means for analyzing voice to generate voice feature data, standard voice feature data storage means for storing standard voice feature data of pre-registered recognizable words, and stored contents of the standard voice feature data storage means Based on, the word detection means for outputting the word detection data for the input speech,
The word detection data from the word detection means is received to understand the meaning of the input voice, and the optimum meaning is selected based on the contents from the state detection means and the conversation content storage means, and the corresponding response content is selected. a speech understanding conversation control means for determining a relative response content decided by the previous SL speech understanding conversation control unit, a voice synthesizing means for generating a speech synthesis output based on the contents of the response data instruction content storage means, this And a voice output unit for outputting the voice synthesis output from the voice synthesis unit to the outside.

The memory means is the age of the speaker or
Consists of a standard voice characteristic data storage means for storing standard voice characteristic data for recognizable words previously registered in correspondence to the sex, etc., the voice recognition response processing unit, a speaker
An audio input means for inputting a voice, a voice analyzing means for generating speech feature data by analyzing the voice input by the voice input unit inputs the speech feature data from the voice analyzing means, said standard voice Based on the stored contents of the characteristic data storage unit, a word detection unit that outputs word detection data for the input voice, and the word detection data from the word detection unit to understand the meaning of the input voice, and the corresponding response content Voice understanding conversation control means, response data instruction content storing means for generating a voice synthesis output based on the response content determined by the voice understanding conversation control means, and the voice understanding conversation control means. And a voice synthesizing means for generating a voice synthesizing output based on the contents of the response data instruction content storing means for the response contents. Characterized in that and an audio output means for outputting the speech synthesized output to the outside from.

Further, the voice input from the speaker is analyzed to generate voice feature data, and this voice feature data is compared with the standard voice feature data of recognizable words registered in advance to obtain word detection data. In a voice recognition dialogue device that outputs, receives the word detection data, understands the meaning of the input voice, determines and outputs the response content corresponding to it, in order to perform voice recognition, the speaker's age or gender, etc. The main body of the apparatus having storage means for storing data preset corresponding to the above, and data preset corresponding to the age or sex of the speaker in order to output a response to the recognized voice. A plurality of cartridges that can be detachably attached to the apparatus main body side are provided by attaching at least one selected from the plurality of cartridges to the apparatus main body. Connected to a provided voice recognition response processing section, the storage means stores standard voice feature data storage means for storing standard voice feature data for recognizable words registered in advance corresponding to the age, sex, etc. of the speaker. And a conversation content storage means for storing response contents corresponding to recognizable words registered in advance corresponding to the age or sex of the speaker, and for instructing what kind of voice synthesis output is generated. And a response data instruction content storage unit for storing instruction content, wherein the voice recognition response processing unit is
State detection means, voice input means for inputting the voice of the speaker, voice analysis means for analyzing the voice input by the voice input means to generate voice characteristic data, and voice characteristic data from the voice analysis means. And a word detection means for outputting word detection data for the input voice based on the stored contents of the standard voice feature data storage means, and understanding the meaning of the input voice by receiving the word detection data from the word detection means. At the same time, based on the contents from the state detecting means and the conversation content storing means, the optimum meaning is selected,
A voice understanding conversation control means for determining a response content corresponding thereto, and a voice synthesis for generating a voice synthesis output based on the content of the response data instruction content storing means with respect to the response content determined by the voice understanding conversation control means. And a voice output unit for outputting the voice synthesis output from the voice synthesis unit to the outside.

[0013]

Then, the voice input from the speaker is analyzed to generate voice feature data, and this voice feature data is compared with the standard voice feature data of recognizable words registered in advance to obtain word detection data. In the voice recognition interactive method of outputting, receiving the word detection data, understanding the meaning of the input voice, determining and outputting the response content corresponding to the input voice, the speaker for outputting the response to the recognized voice. preset data so as to correspond to the age or sex, etc.
By providing a plurality of cartridges detachably attachable to the apparatus main body having a storage means for storing, and attaching at least one selected from the plurality of cartridges to the apparatus main body, The stored contents connected to the provided voice recognition response processing unit and stored in the storage means include response contents corresponding to recognizable words registered in advance corresponding to the age or sex of the speaker. It indicates whether to generate a speech synthesis output as formed by the response data instruction content, the speech recognition response processing unit, the state
Based on standard voice feature data of pre-registered recognizable words, a state detection process of detecting a voice, a voice analysis process of analyzing voice input by voice input means to generate voice feature data, for a word detection step of outputting a word detection data, as well as understand the meaning of the input voice receiving word detection data from the word detection process, before
The state detected by the state detection step and the content of the response
Based on bets, select the best sense, determines the response content corresponding thereto, and speech understanding conversation control step of determining with reference to the stored conversation in the storage means side, by the speech understanding conversation control process A voice synthesizing step for generating a voice synthesizing output based on the response data instruction content stored in the storage means side, and a voice outputting step for outputting the voice synthesizing output from the voice synthesizing step to the outside. It is characterized by having.

Also, the voice input from the speaker is analyzed to generate voice feature data, and this voice feature data is compared with the standard voice feature data of recognizable words registered in advance to obtain word detection data. In the voice recognition interactive method of outputting, receiving the word detection data, understanding the meaning of the input voice, determining and outputting the corresponding response content, in order to perform voice recognition, the age or sex of the speaker, etc. The main body of the apparatus having storage means for storing data preset corresponding to the above, and data preset corresponding to the age or sex of the speaker in order to output a response to the recognized voice. A plurality of cartridges that can be detachably attached to the apparatus main body side are provided by attaching at least one selected from the plurality of cartridges to the apparatus main body. Connected to the provided voice recognition response processing unit, the stored contents stored in the storage means are standard voice feature data for recognizable words registered in advance corresponding to the age or sex of the speaker, the speaker Formed of response contents corresponding to pre-registered recognizable words corresponding to the age, sex, etc., and response data instruction contents storing instruction contents for instructing what kind of speech synthesis output is generated Then, the voice recognition response processing unit detects the state.
A state detecting step, a voice analyzing step of analyzing voice input by the voice input means to generate voice characteristic data,
The word detection step of inputting the voice feature data from this voice analysis step and outputting the word detection data for the input voice based on the stored contents stored in the storage means, and the word detection data from this word detection step In addition to understanding the meaning of the input voice, it is detected by the state detection step.
A voice understanding conversation control step of selecting an optimum meaning based on the state and the response content, and determining the corresponding response content by referring to the conversation content stored in the storage means; A speech synthesis step for generating a speech synthesis output based on the response data instruction content stored in the storage means for the response content determined by the understanding conversation control step, and the speech synthesis output from this speech synthesis step to the outside. And a voice output step of outputting.

[0016] Then, the storage contents stored in said storage means is a standard voice characteristic data stored standard voice characteristic data for the speaker age or recognizable words previously registered in correspondence to the sex, etc. , The voice recognition response
The processing unit, and the speech analysis process of generating a speech feature data by analyzing the voice inputted by the voice input unit inputs the speech feature data from the voice analysis step, storing the standard voice characteristic data storage means A word detection process that outputs word detection data for the input voice based on the content, and a voice understanding conversation that receives the word detection data from this word detection process to understand the meaning of the input voice and determines the corresponding response content. A control step, a voice synthesizing step for performing voice synthesizing based on the response data instruction content to instruct what kind of voice synthesizing output should be given to the response content determined by the voice understanding conversation control step, and this voice synthesizing step And a voice output step of outputting a voice synthesis output from the outside to the outside.

The memory contents stored in the memory means include response contents corresponding to recognizable words registered in advance corresponding to the age or sex of the speaker, and what kind of voice synthesis output is generated. a response data instruction contents instructing whether the recognition, wherein the speech recognition response processing unit, a speech analysis process of generating a speech feature data by analyzing the voice inputted by the voice input means, which is registered in advance Based on the standard speech feature data of possible words, the word detection process that outputs the word detection data for the input voice, and the word detection data from this word detection process is received to understand the meaning of the input voice and the corresponding response With respect to the voice comprehension conversation control step of determining the content by referring to the conversation content stored in the storage means side, and the storage with respect to the content determined by the voice comprehension conversation control step A voice synthesizing step for generating a voice synthesizing output based on the response data instruction content stored on the column side, and a voice outputting step for externally outputting the voice synthesizing output from this voice synthesizing step .

[0018]

The stored contents stored in the storage means are standard voice feature data for recognizable words registered in advance corresponding to the age or sex of the speaker , the speaker
Response data instruction content that stores response content corresponding to a recognizable word registered in advance corresponding to the age, gender, etc., and instruction content for instructing what kind of voice synthesis output is generated. The voice recognition response processing unit receives a voice analysis step of analyzing voice input by a voice input unit to generate voice feature data, and inputs voice feature data from the voice analysis step. Based on the stored contents stored in the storage means, a word detection step of outputting word detection data for the input voice, and receiving the word detection data from this word detection step, understanding the meaning of the input voice, and responding thereto. With respect to the voice understanding conversation control step of determining the response content with reference to the conversation content stored in the storage means, and the storage of the response content determined by the voice understanding conversation control step. A speech synthesis step of generating a speech synthesis output based on the response data instruction content stored in the stage, characterized in that and an audio output step of outputting a speech synthesis output from the speech synthesis step to the outside.

[0020]

According to the present invention, the preset memory contents for voice recognition and the preset memory contents for outputting a response to the recognized voice can be detachably attached to the main body of the apparatus. Since it is stored in the cartridge side and this cartridge is mounted in the main body of the device, the response content to the input voice is output based on the data stored in the cartridge, so that the main body of the device is one unit. Even if there is one, changing the cartridge makes it possible to have conversations according to various ages or genders.
Therefore, the cartridge can be selected according to the user, and the recognizable word and the response content can be selected for each cartridge. Therefore, it is possible to give a wide variation to the content of the dialogue and the like. It will be possible to operate

[0021]

Embodiments of the present invention will be described below with reference to the drawings. In this embodiment, the case where the present invention is applied to a toy is taken as an example, and here, the case where the present invention is applied to a stuffed animal such as a dog will be described. Further, an example in which the present invention is applied to an unspecified speaker voice recognition device capable of recognizing an unspecified speaker's voice will be described.

(First Embodiment) FIG. 1 is a diagram for explaining the overall schematic configuration of the present invention. In general, a voice recognition response process stored in a stuffed dog (device body) 1 is shown. It is composed of a section 10 (details will be described later) and a cartridge section 20 (details will be described later) detachably mountable to a predetermined portion of the stuffed dog 1.

FIG. 2 is a block diagram for explaining the configurations of the voice recognition response processing section 10 and the cartridge section 20 according to the first embodiment. In the first embodiment, an example will be described in which three ROMs including a standard voice feature data storage unit 21, a conversation content storage unit 22, and a response data instruction content storage unit 23 are provided on the cartridge side.

The voice recognition response processing means 10 is composed of a voice input unit 11, a voice analysis unit 21, a word detection unit 13, a voice understanding conversation control unit 14, a voice synthesis unit 15, a voice output unit 16 and the like. Note that among these components, the voice analysis unit 12, the word detection unit 13, the voice understanding conversation control unit 1
4. The voice synthesizing unit 15 and the like are stored in the stuffed toy 1, for example, near the abdomen, the voice input unit (microphone) 11 is provided, for example, in the ear part of the stuffed toy 1, and the voice output unit (speaker) 16 is provided in the mouth, for example. .

On the other hand, the cartridge unit 20 is composed of a standard voice feature data storage unit 21, a conversation content storage unit 22, and a response data instruction content storage unit 23, and a cartridge mounting unit provided near the abdomen of the stuffed toy 1 (see FIG. It is easily removable from the outside (not shown). When the cartridge unit 20 is properly mounted in the cartridge mounting unit, it is connected to each unit of the voice recognition response processing unit 10 and signals can be transmitted and received. Specifically, the standard voice feature data storage unit 21 is connected to the word detection unit 13, the conversation content storage unit 22 is connected to the voice understanding conversation control unit 14, and the response data instruction content storage unit 23 is the voice understanding. It is adapted to be connected to the conversation control unit 14 and the voice synthesis unit 15.

The standard voice feature data storage unit 21 stores 1
This is a ROM that stores (registers) a standard pattern of recognizable words (referred to as registered words) created in advance using voices spoken by a large number of speakers (for example, about 200 people) for each word. . In this case, because it is a stuffed toy as an example, the registration word is about 10 words, as is the word, for example, "Good morning", "good night", "Hello", used to greeting such as "tomorrow", "weather" words However, the number of registered words is not limited to 10 and the number of registered words is not limited to 10. Further, the conversation content storage unit 22 stores contents such as what kind of word is a registered word and what kind of response is made to each registered word. This conversation content storage unit 22 is
Originally, it is provided in the voice comprehension / conversation control unit 14, but it is provided on the cartridge unit 20 side because the registered word may change depending on the cartridge. Further, the response data instruction content storage unit 23 stores the content instructing what kind of voice synthesis output should be performed corresponding to each registered word. Even if the response content is the same, for example, depending on how the boy speaks. The contents mainly instructing the voice quality are preset, such as voice synthesis output or voice synthesis output according to the way the girl speaks.

By the way, as described above, the cartridge section 20 is prepared with various kinds of standard voice feature data, registered words, further, response contents to each registered word, voice quality, and the like. A user who uses the stuffed toy 1 can arbitrarily select it. For example, in the cartridge for infants, a plurality of registered words based on the standard voice feature data that makes it easy to recognize the voice of the infant is stored in the standard voice feature data storage unit 21, and what kind of word is the registered word. Conversation content storage unit 22 determines how to respond to each registered word.
Further, the response data instruction content storage unit 23 stores instructions such as what kind of voice synthesis output should be made for each registered word.

Next, the respective functions of the above-mentioned respective parts, and the overall processing will be described below in order.

The voice input unit 11 of the voice recognition response processing unit 10
Is composed of a microphone, an amplifier, a low-pass filter, an A / D converter, etc., which are not shown in the figure. The sound input from the microphone is converted into an appropriate sound waveform through the amplifier and the low-pass filter, and then by the A / D converter. Digital signal (for example, 12KHz.16bit)
And outputs the converted output to the voice analysis unit 12.
In the voice analysis unit 12, the voice waveform signal sent from the voice input unit 11 is subjected to frequency analysis for each short time using a computing unit (CPU), and a multidimensional feature vector representing a feature of frequency is extracted ( The LPC-CEPSTRUM coefficient is common), and the time series of this feature vector (hereinafter referred to as a voice feature vector sequence) is output.

Although not shown in the figure, the word detecting unit 13 mainly stores an arithmetic unit (CPU) and a processing program.
It is configured to detect which part of the input voice the word, which is composed of M and is registered in the standard voice feature data storage unit 21 provided in the cartridge unit 20, exists and with certainty. As the word detecting unit 13,
It is also possible to use a hidden Markov model (HMM) method, a DP matching method, or the like.
Unspecified using the keyword spotting processing technology by the N (dynamic recurrent neural network) method (for this technology, the applicant has already applied for a patent according to Japanese Patent Application Laid-Open Nos. 6-4097 and 6-119476). It is assumed that the word detection data is output so as to enable speech recognition similar to continuous speech recognition by a speaker.

The specific processing of the word detection unit 13 will be briefly described with reference to FIG. Word detector 1
3 detects the word registered in the standard voice feature data storage unit 21 and the certainty of which part of the input voice exists. Now, the speaker inputs a voice such as "Tomorrow's weather is ..."
It is assumed that an audio signal as shown in (a) is output. In the phrase "weather of tomorrow ...", "tomorrow" and "weather" are the keywords in this case, and these are standard words as one of the registered words of about 10 words in advance. The pattern is stored in the voice feature data storage unit 21. Then, these registered words are, for example, 10
When the word is used, a signal for detecting each word is output corresponding to these 10 words (this is referred to as word 1, word 2, word 3, ...), and the value of the detection signal is output. From such information, it is detected with certainty that the corresponding word is present in the input voice. That is, when the word "weather" (word 1) is present in the input voice, the detection signal waiting for the signal "weather" indicates the "weather" of the input voice as shown in FIG. Get up in parts. Similarly, when the word "tomorrow" (word 2) is present in the input voice, the detection signal waiting for the signal "tomorrow" is "tomorrow" of the input voice as shown in FIG. Stand up at the part. In FIGS. 9B and 9C, the numerical value such as 0.9 or 0.8 is a numerical value indicating the certainty (approximation degree), and if the numerical value is high such as 0.9 or 0.8, the registered word having the high certainty is , It can be said that the input voice is a recognition candidate.
That is, the registered word "tomorrow" exists at a certainty of 0.8 in the portion of w1 on the time axis of the input voice signal, as shown in FIG.
As shown in FIG. 7B, w on the time axis of the input audio signal
It can be seen that there is a certainty of 0.9 in the 2nd part.

In the example of FIG. 3, in response to the input of "weather", as shown in FIG. 3D, the word 3 (this word 3 is a registered word "what time") is input. The waiting signal also rises with a certain degree of certainty (its numerical value is about 0.6) in the portion of w2 on the time axis.
Thus, when two or more registered words are present as recognition candidates at the same time with respect to the input voice signal, the word having the highest degree of approximation (a numerical value indicating the certainty) is selected as the recognition word. A method, a correlation table representing a correlation rule between each word is created in advance, and a method of selecting any one word as a recognition word by this correlation table is used to identify a certain recognition candidate word. decide. For example, if we decide the recognition candidates using the former method, in this case,
Since the approximation degree corresponding to the portion of w2 on the time axis has the highest approximation degree of the detection signal for detecting "weather", it is determined that the recognition candidate for the input voice of that portion is "weather". . The voice understanding conversation control unit 14 recognizes the input voice based on these degrees of approximation.

The speech comprehension conversation control section 14 is mainly composed of an arithmetic unit (CPU) and a ROM storing a processing program, inputs the word detection data from the word detection section 13, and outputs the word detection data. Based on this, the voice is recognized (the meaning of the entire input voice is understood), and the conversation content storage unit 22 provided in the cartridge unit 20 is referenced to determine the response content according to the meaning of the input voice and With reference to the data instruction content storage unit 23, a signal indicating what kind of voice synthesis output is to be sent to the voice synthesis unit (mainly composed of a CPU and ROM) 15.

For example, FIG. 3 from the word detection unit 13
Detection data as shown in (b) to (e) (this is called word lattice. This word lattice is a registered word name,
(Including the degree of approximation, a signal indicating the start point s and the end point e of a word, etc.) is input, first, based on the word lattice, one or more words as keywords in the input voice are determined. . In this example, since the input voice is "Tomorrow's weather is ...", "Tomorrow" and "Weather" will be detected, and "Tomorrow's weather" will be detected from the keywords of "Tomorrow" and "Weather". Understands the content of continuous input voice such as "...", and selects and outputs the response content corresponding to it. In addition, in this case, the response content is such that "the weather of tomorrow is sunny".
State detection means (temperature detection unit, not shown here)
For example, if it is information about the weather, the change in the weather is judged based on the change in the atmospheric pressure from the atmospheric pressure detection unit, and the atmospheric pressure tends to increase. If so, the response content corresponding thereto is read from the response data instruction content storage unit 23. Similarly, responses regarding temperature, time, date, etc. are also possible.

The speech recognition processing similar to the continuous speech recognition by the keyword spotting processing described above can be applied not only to Japanese but also to other languages. For example, if your language is English, the recognizable words that are registered are, for example, “good mornin
“G”, “time”, “tommorow”, “good night”, etc. are given as examples, and the feature data of the recognizable registered words are stored in the standard voice feature data storage unit 21. And now, Person asked "what time is it now", the word "time" was the keyword in this case in the phrase "what time is it now", and the word "time" was present in the input speech. Sometimes, the detection signal waiting for that "time" voice signal,
Start up at the "time" part of the input voice. Then, when the detection data (word lattice) from the word detection unit 13 is input, first, one or a plurality of words as keywords in the input voice are determined based on the word lattice. In this example, the input voice is "what time is it
Since it is now, "time" will be detected as a keyword, and based on this keyword, "what ti
Understand the contents of continuous input speech "me is it now".

The CPUs for controlling the above-mentioned voice analysis, word detection, voice comprehension conversation control, voice synthesis and the like may be provided respectively, but one main CPU for performing all of these processes is provided, The entire processing of the present invention may be performed by this one CPU.

In such a configuration, if the mounted cartridge is intended for an infant, for example, if the infant asks "Good morning", the input voice is analyzed by the voice analysis unit 12. Then, it is sent to the word detection unit 13. Then, in the word detection unit 13, based on the feature data stored in the standard voice feature data storage unit 21, the word detection unit 13 performs the above-described processing, and word detection data (word lattice) for the input voice is obtained. Is output. At this time, the cartridge unit 20
Since it is for infants, the contents of the standard voice feature data storage unit 21 are standard voice feature data obtained based on the features of the voice of the infant, and therefore recognition can be performed at a high recognition rate.

Then, the voice understanding conversation control unit 14, which has received the word lattice from the word detecting unit 13, refers to the conversation content storing unit 22 of the cartridge unit 20 based on the word lattice and the input voice is "Good morning. , And after getting the response content to it,
The response data instruction content storage unit 23 is referred to. As a result, in the voice synthesis unit 15, the response data instruction content storage unit 2
A voice synthesis output is produced based on the information obtained from 3, and the voice output unit 16 outputs the response contents. In this case, the content of the response is a response to the infant, and therefore, for example, "good morning" is answered in the way of speaking for the infant.

In this way, by mounting the cartridge unit 20 arbitrarily selected by the user on the stuffed animal 1, it becomes possible to have a dialogue according to the contents of the cartridge unit 20. For example, if a cartridge for infants as described above is attached, dialogue for infants can be performed, and if a cartridge for elementary school children is attached, dialogue according to it can be performed. It is possible to prepare various cartridges according to various ages or sexes. Specifically, it is possible to perform detailed classifications for boys and toddlers, girls, toddlers in elementary school, and girls.

As a result, even if the apparatus main body (plush toy in this case) is only one, by changing the cartridge, it is possible to have a dialogue according to various ages or sexes. Further, in this case, since the standard voice feature data storage unit 21, the conversation content storage unit 22, and the response data instruction content storage unit 23 are incorporated in the cartridge unit 20, the recognizable words and the standard voice feature data thereof are stored in the cartridge. It can be different for each cartridge, and the response content and voice quality for the recognizable word can be different for each cartridge. Therefore,
By increasing the variety of cartridge types, it becomes possible to have dialogues according to various ages or genders.

(Second Embodiment) In the first embodiment described above, an example in which the standard voice feature data storage unit 21, the conversation content storage unit 22, and the response data instruction content storage unit 23 are provided in the cartridge unit 20 is described. As described above, the cartridge section 20 may be provided with only the standard voice feature data storage section 21 in the cartridge section 20 without having all of these three elements, and the response data instruction content storage section may be provided. Only 23 may be provided in the cartridge portion 20, and various combinations thereof are conceivable.

Here, as the cartridge unit 20, only the standard voice characteristic data storage unit 21 is used.
If the conversation content storage unit 22 and the response data instruction content storage unit 23 are provided in the cartridge unit 20,
Only the response data instruction content storage unit 23 is used as the cartridge unit 2.
The case where it is provided within 0 will be described respectively.

First, the case where only the standard voice feature data storage unit 21 is provided in the cartridge unit 20 will be described. FIG. 4 is a block diagram showing the configuration, and the same parts as those in FIG. 2 are designated by the same reference numerals. That is, in this case, the conversation content storage unit 22 and the response data instruction content storage unit 23 are provided on the voice recognition response processing unit 10 side.
The response content storage unit 22 may be provided in the voice comprehension conversation control unit 14, but here it is shown that it is provided separately.

In this way, the standard voice feature data storage unit 21
Only the cartridge portion 20 is provided, and the conversation content storage unit 2 is provided.
When 2 and the response data instruction content storage unit 23 are on the device (stuffed toy 1) side, the device side determines in advance how to respond to each registered word. Therefore, in this case, the registration word, word predetermined by apparatus (e.g., as described above, "good morning", "Hello", 10 about words like that as "Good night") but only, respectively It is possible to prepare various standard voice feature data for each cartridge according to the age and sex of the registered word. For example, a cartridge storing standard voice feature data of infants for each registered word, a cartridge storing standard voice feature data of elementary school students for each registered word,
A cartridge that stores standard voice feature data of an adult female for each registered word, a cartridge that stores standard voice feature data of an adult male for each registered word, such as the age and gender. A cartridge storing standard voice feature data for each registered word is prepared for each item.

Thus, for example, when used by an infant, by selecting a cartridge in which voice feature data for the infant is stored and mounting it in the main body of the apparatus, the features of the voice of the infant are analyzed. Since it can be compared with the obtained voice feature data, the recognition can be performed with a high recognition rate. Then, a predetermined response content is speech-synthesized and output for the recognized word. In this way, the recognition rate can be significantly improved by selecting the cartridge of the standard voice feature data according to the age and sex.

Next, a case where the conversation content storage unit 22 and the response data instruction content storage unit 23 are provided in the cartridge unit 20 will be described. FIG. 5 is a block diagram showing the configuration, and the same parts as those in FIG. 2 are designated by the same reference numerals.
That is, in this case, the standard voice feature data storage unit 21
Is provided on the voice recognition response processing unit 10 side, and the conversation content storage unit 22 and the response data storage unit 23 are provided on the cartridge unit 20 side.

As described above, when only the conversation content storage unit 22 and the response data instruction content storage unit 23 are provided on the cartridge unit 20 side and the standard voice feature data storage unit 21 is provided on the voice recognition response processing unit 10 side, registration is performed. The word is a predetermined word on the device side (for example, "Good morning" as described above,
"Hello", "Good night," but only 10 about a word) such that like, be variously prepared speech synthesis output to the response content and each of the response content for each of the registered words (such as voice quality) for each cartridge You can For example, what kind of response contents should be given to the word "Good morning", and what kind of voice synthesis output the response contents should be, should be decided in advance for each cartridge. It is stored in the conversation content storage unit 22 and the response data instruction content storage unit 23. Specifically, the cartridge for infants is a cartridge for elementary school students that responds to registered words such as "Good morning" and "Good night" with the response content for infants and the sound quality for infants. As for the response to the registered word, for example, by imitating the way the character of the TV animation is spoken and responding with the response content for elementary school students, the response content and voice of each age and gender are matched. Prepare various cartridges that respond by quality.

In this way, for example, when the elementary school student uses the cartridge for elementary school students as described above and attaches it to the device, the elementary school student speaks some registered word. It becomes possible to return the response contents set in the cartridge with the quality of the voice imitating the talking style of the character of the TV animation.

Next, a case where only the response data instruction content storage unit 23 is provided on the cartridge unit 20 side will be described. FIG. 6 is a block diagram showing the configuration, and the same parts as those in FIG. 2 are designated by the same reference numerals. That is, in this case, the standard voice feature data storage unit 21 and the conversation content storage unit 22 are provided on the voice recognition response processing unit 10 side of the apparatus body, and only the response data instruction content storage unit 23 is provided on the cartridge unit 20 side. ing.

In this way, the response data instruction content storage unit 23
Only the cartridge portion 20 is provided, and the conversation content storage unit 2 is provided.
2 and the standard voice feature data storage unit 21 are provided on the voice recognition response processing unit 10 side, the registered word is a word predetermined by the device side (for example, “Ohayo” as described above,
"Hello", "Good night" is only 10 about a word) such as such, also, the response content to these registered words is basically the content which is previously set on the device side, for that response content It is possible to set various types of voice synthesis output for each cartridge.
For example, it is determined in advance what kind of voice quality the response content to the word "Ohayo" is to be output, and these are stored in the response data instruction content storage unit 23 for each cartridge. Specifically, in the case of cartridges for infants, "Good morning" and "Good night"
For various registered words such as, respond to those registered words with an instruction to make a voice response like a mother, or to make a response that resembles the voice of a TV anime character for young children. As a cartridge for elementary school students, the instruction content such as responding to the registered word by simulating the way the character of the TV animation character for elementary school students speaks is stored.
Various cartridges are prepared for each registered word, in which contents for instructing what kind of voice synthesis output is to be stored are prepared for each age and sex. At this time,
The contents of each response to each registered word are basically preset contents.

In this way, for example, when the elementary school student uses the cartridge for elementary school students as described above and mounts it on the device, when the elementary school student speaks some registered word, the response contents Is basically a preset content, but it is possible to receive a response with a voice imitating the way of speaking the character of the TV animation for elementary school students as described above.

As described above, in the present invention, the ROM portion including the standard voice feature data storage unit 21, the conversation content storage unit 22, the response data instruction content storage unit 23, etc. is of a cartridge type, and can be used for the age and sex of the user. Various kinds of cartridges having standard voice characteristic data and response contents are prepared so that the user can arbitrarily select them. Therefore, even if the recognition response device itself is one, by replacing the cartridge, it can be used by a wide range of people, and voice recognition and dialogue can be performed according to the user.

In each of the above embodiments, the present invention is applied to a plush toy as a toy, but the present invention is not limited to a plush toy. It can be applied not only to other toys, but also to toys, as well as game machines and various electronic devices used in daily life, and its application range is considered to be extremely wide.

[0054]

As described above, in the voice recognition dialogue apparatus of the present invention, a storage means for storing preset memory contents for voice recognition and response to the recognized voice is provided to the apparatus body. Since it is provided on the side of the cartridge that can be detachably mounted, and the cartridge is mounted on the side of the main body of the device, the response content to the input voice is output based on the data stored in the cartridge. Even if there is only one main body of the apparatus, by changing the cartridge, it becomes possible to have a dialogue according to various ages or sexes. Therefore, when the present invention is applied to, for example, a toy, a cartridge can be selected according to the growth of children, and recognizable words and response contents can be variously selected by selecting the cartridge. Since it is possible, even one toy does not get tired on the way, and it can be used for a long period of time, and it can be widely used regardless of age or sex. Furthermore, not only toys but also when applied to electronic devices and the like, by selecting a cartridge that is suitable for the user, it is possible to give a wide variety of dialogue contents, and to perform various actions corresponding to it. The effect is extremely large.

Further, since the standard voice feature data storage means for storing the standard voice feature data of pre-registered recognizable words is provided on the cartridge side, the standard voice corresponding to the age and sex of each registered word is provided. Various characteristic data can be prepared for each cartridge, and a recognition rate can be significantly improved by selecting and using a cartridge suitable for the user.

Further, conversation content storage means for storing response contents corresponding to pre-registered recognizable words, and response data instruction content storage for storing instruction contents for instructing what kind of voice synthesis output should be generated. Since the means is provided on the cartridge side, it is possible to prepare various cartridges having response contents according to each age and sex. For example, if a child-friendly cartridge is selected, if some registered word is spoken, , It becomes possible to respond with a response content for children and a voice for children.

Further, since the response data instruction content storage means for storing the instruction content for instructing what kind of voice synthesis output is generated is provided on the cartridge side, it is possible to match the age and sex of each with For each registered word,
It is possible to prepare various cartridges that store the content that dictates what kind of voice synthesis output is used. For example, if you select a cartridge for elementary school students, when you speak some registered word, the response content is basically Is a preset content, but it is possible to receive a response with a voice imitating the way the character of the television animation character for elementary school students as described above is imitated.

Also, standard voice feature data storage means for storing standard voice feature data of pre-registered recognizable words, and conversation content storage means for storing response contents corresponding to the registered recognizable words. Since the response data instruction content storage means for instructing what kind of voice synthesis output is generated is provided on the cartridge side, the standard voice feature data corresponding to the age and gender of each registered word is provided for each cartridge. Various types of cartridges can be prepared, and the recognition rate can be significantly improved by selecting and using a cartridge that is suitable for the user. In addition, recognizable registered words can be set for each cartridge, which allows for variations in conversation. Can be significantly increased. Further, various cartridges having response contents and voice synthesis output suitable for each age and sex can be prepared. As a result, even in the case of one apparatus main body, by changing the cartridge, it becomes possible to have a wide variety of dialogues according to various ages or sexes.

The speech recognition dialogue processing method of the present invention is
By providing a preset memory content for voice recognition and a response to the recognized voice on the cartridge side that is detachably mountable to the apparatus body, and by mounting this cartridge on the apparatus body side, Based on the data stored in the cartridge, the response content to the input voice is generated. Therefore, even if there is only one main body of the device, the cartridge can be changed to meet various ages or sexes. Dialogue becomes possible. Therefore, when the present invention is applied to, for example, a toy,
You can choose a cartridge that suits your child's growth, and various recognizable words and responses can be selected by choosing a cartridge, so you will not get bored with one toy. It can be used for a long period of time, and it can be used widely regardless of age or sex. Furthermore, not only toys but also when applied to electronic devices and the like, by selecting a cartridge that is suitable for the user, it is possible to give a wide variety of dialogue contents, and to perform various actions corresponding to it. The scope of application is extremely wide.

Further, since the standard voice characteristic data of the pre-registered recognizable words are stored in the cartridge side, various standard voice characteristic data corresponding to the age and sex of each registered word are prepared for each cartridge. By selecting and using a cartridge suitable for the user, the recognition rate can be significantly improved.

Further, since the contents of the response corresponding to the pre-registered recognizable word and the contents of the instruction for instructing what kind of voice synthesis output should be generated are stored in the cartridge side, each age and It is possible to prepare various cartridges with response contents tailored to gender, for example, if you select a cartridge for children,
Speaking some registered word makes it possible to respond with a response content for children and a voice for children.

Further, since the contents of the instruction for instructing what kind of voice synthesis output is to be generated are stored in the cartridge side, each registered word can be stored according to each age and sex. It is possible to prepare various cartridges that store the content that dictates what kind of voice synthesis output is used. For example, if you select a cartridge for elementary school students, when you speak some registered word, the response content is basically Is a preset content, but it is possible to receive a response with a voice imitating the way the character of the television animation character for elementary school students as described above is imitated.

Standard voice feature data of recognizable words registered in advance, response contents corresponding to the registered recognizable words, response data instruction contents for instructing what kind of voice synthesis output is to be generated. Since it is stored in the cartridge side, various standard voice feature data can be prepared for each cartridge according to the age and sex of each registered word, and the cartridge suitable for the user can be selected and used. As a result, the recognition rate can be significantly improved, and recognizable registered words can also be set for each cartridge, so that the variation in conversation can be greatly increased. Further, various cartridges having response contents and voice synthesis output suitable for each age and sex can be prepared. This gives 1
Even with the main body of the stand, by changing the cartridge, it is possible to have a wide variety of dialogues according to various ages or sexes.

[Brief description of drawings]

FIG. 1 is a diagram illustrating an outline of the present invention.

FIG. 2 is a block diagram illustrating a first embodiment of the present invention.

FIG. 3 is a diagram illustrating a word detection process by a word detection unit and a voice recognition process by a voice understanding conversation control unit.

FIG. 4 is a block diagram illustrating a second embodiment (No. 1) of the present invention.

FIG. 5 is a block diagram illustrating a second embodiment (No. 2) of the present invention.

FIG. 6 is a block diagram illustrating a second embodiment (No. 3) of the present invention.

[Explanation of symbols]

1 ... Plush toy (device body) 10 ... Voice recognition response processing unit 11 ... Voice input section 12 ... Voice analysis unit 13 ... Word detection unit 14 ... Voice understanding conversation control unit 15 ... Speech synthesizer 16 ... Voice output section 20 ... Cartridge part 21 ... Standard voice feature data storage unit 22 ... Conversation content storage section 23 ... Response data instruction content storage unit

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Hiroshi Hasegawa               3-5 Yamato 3-chome, Suwa City, Nagano Prefecture               Within Co-Epson Corporation (72) Inventor Isao Tsuneo               3-5 Yamato 3-chome, Suwa City, Nagano Prefecture               Within Co-Epson Corporation (72) Inventor Osamu Urano               3-5 Yamato 3-chome, Suwa City, Nagano Prefecture               Within Co-Epson Corporation                (56) Reference literature JP-A-6-133039 (JP, A)                 JP-A-4-93899 (JP, A)                 Japanese Patent Laid-Open No. 4-167071 (JP, A)

Claims (4)

(57) [Claims] 1. A voice input from a speaker is analyzed to generate voice feature data, and this voice feature data is compared with standard voice feature data of recognizable words registered in advance to obtain word detection data. and outputs, in response to this word detection data, to understand the meaning of the input voice, the voice recognition interaction device for determining and outputting the response content corresponding thereto, the talk in order to perform a response output to recognized speech Data set in advance according to the age or gender of the person
A storage means for storing data in a detachably plurality of cartridges which can be mounted is provided to the apparatus main body, by at least one selected from the plurality of cartridges attached to the main body, the apparatus body side Connected to a voice recognition response processing unit provided in, the storage means, a conversation content storage means for storing response content corresponding to a recognizable word registered in advance corresponding to the age or gender of the speaker, , A response data instruction content storage unit that stores instruction content for instructing what kind of voice synthesis output is to be generated, the voice recognition response processing unit includes a state detection unit and a speaker voice. A voice input means for inputting, a voice analyzing means for analyzing voice input by the voice input means to generate voice feature data, and a standard voice feature of a pre-registered recognizable word. A standard voice feature data storage unit that stores data, a word detection unit that outputs word detection data for an input voice based on the stored contents of the standard voice feature data storage unit, and a word detection data from this word detection unit. Understand the meaning of the input voice and detect the status
Based on the content from the means and the conversation content storage means,
Select the best sense, a speech understanding conversation control means for determining a response content corresponding thereto, relative to the response content decided by the previous SL speech understanding conversation control means, based on the contents of the response data instruction content storage means A voice recognition dialogue apparatus comprising: a voice synthesizing means for generating a voice synthesizing output; and a voice outputting means for outputting the voice synthesizing output from the voice synthesizing means to the outside. 2. A voice input from a speaker is analyzed to generate voice feature data, and the voice feature data is compared with standard voice feature data of recognizable words registered in advance to obtain word detection data. In a voice recognition dialogue device that outputs, receives the word detection data, understands the meaning of the input voice, determines and outputs the corresponding response content, in order to perform voice recognition, the age or sex of the speaker, etc. The main body of the apparatus having storage means for storing data preset corresponding to the above, and data preset corresponding to the age or sex of the speaker in order to output a response to the recognized voice. A plurality of cartridges detachably attachable to the device main body are provided, and at least one selected from the plurality of cartridges is attached to the device main body to install the cartridge on the device main body side. Connected to a voice recognition response processing unit, the storage means, a standard voice feature data storage means for storing standard voice feature data for recognizable words registered in advance corresponding to the age or sex of the speaker, , Conversation content storage means for storing response contents corresponding to recognizable words registered in advance corresponding to the age or sex of the speaker, and instructions for instructing what kind of speech synthesis output is generated The voice recognition response processing unit includes a state detecting unit, a voice input unit for inputting a voice of a speaker, and a voice input by the voice input unit. Voice analysis means for analyzing and generating voice feature data, and voice feature data from this voice analysis means are input, and based on the stored contents of the standard voice feature data storage means, the input sound is input. A word detection unit for outputting a word detection data, to understand the meaning of the input voice receiving word detection data from the word detection means with respect to the state
Based on the contents from the detection means and the conversation content storage means, a voice understanding conversation control means for selecting an optimum meaning and determining a response content corresponding thereto, and a response content determined by the voice understanding conversation control means On the other hand, it is characterized by having a voice synthesizing means for generating a voice synthesizing output based on the contents of the response data instruction content storing means, and a voice outputting means for outputting the voice synthesizing output from the voice synthesizing means to the outside. Speech recognition interactive device. 3. A voice input from a speaker is analyzed to generate voice feature data, and this voice feature data is compared with standard voice feature data of recognizable words registered in advance to obtain word detection data. In the voice recognition interactive method of outputting, receiving the word detection data, understanding the meaning of the input voice, determining and outputting the corresponding response content, in order to output the response to the recognized voice, the speaker The date set in advance according to the age, gender, etc.
A storage means for storing data in a detachably plurality of cartridges which can be mounted is provided to the apparatus main body, by at least one selected from the plurality of cartridges attached to the main body, the apparatus body side Connected to a voice recognition response processing unit provided in, the storage content stored in the storage means, the response content corresponding to a recognizable word pre-registered in association with the age or gender of the speaker, what is formed by the response data instruction content indicates whether to generate a speech synthesis output, wherein the voice recognition response processing unit, the state detection for detecting the state
Based on the step, a voice analysis step of analyzing the voice input by the voice input means to generate voice feature data, and the standard voice feature data of recognizable words registered in advance, the word detection data for the input voice is obtained. The word detection step for outputting and the word detection data from this word detection step are received to understand the meaning of the input voice, and the state detection step is performed.
Based on the more detected state and the response content ,
For the speech understanding conversation control step of determining the optimum meaning and determining the corresponding response content with reference to the conversation content stored in the storage means side, and the content determined by this speech understanding conversation control step A voice synthesizing step for generating a voice synthesizing output based on the response data instruction content stored in the storage means side, and a voice outputting step for outputting the voice synthesizing output from the voice synthesizing step to the outside. Characteristic speech recognition dialogue processing method. 4. The speech input data from a speaker is analyzed to generate speech feature data, and this speech feature data is compared with standard speech feature data of recognizable words registered in advance to determine word detection data. In the voice recognition dialogue method of outputting, receiving the word detection data, understanding the meaning of the input voice, determining and outputting the corresponding response content, in order to perform voice recognition, the age or gender of the speaker, etc. The main body of the apparatus having storage means for storing data preset corresponding to the above, and data preset corresponding to the age or sex of the speaker in order to output a response to the recognized voice. A plurality of cartridges detachably attachable to the device main body are provided, and at least one selected from the plurality of cartridges is attached to the device main body to install the cartridge on the device main body side. Connected to the voice recognition response processing unit, the storage content stored in the storage means, the standard voice feature data for a recognizable word registered in advance corresponding to the age or gender of the speaker, It is formed of response contents corresponding to recognizable words registered in advance corresponding to the age, sex, etc., and response data instruction contents storing instruction contents for instructing what kind of voice synthesis output is generated. The voice recognition response processing unit includes a state detection unit that detects a state.
A voice analysis step of analyzing voice input by the voice input means to generate voice feature data, and voice feature data from the voice analysis step, and based on the stored contents stored in the storage means. In the word detection step of outputting the word detection data for the input voice, while understanding the meaning of the input voice by receiving the word detection data from the word detection step , the state detected by the state detection step,
A voice understanding conversation control step of selecting an optimum meaning based on the response content and determining the corresponding response content with reference to the conversation content stored in the storage means, and this voice understanding conversation control step. A voice synthesis step for generating a voice synthesis output based on the response data instruction content stored in the storage means, and a voice output for externally outputting the voice synthesis output from the voice synthesis step in response to the response content determined by And a voice recognition dialogue processing method.