JP2001022384A - Voice interactive device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce an unpleasant feeling and an uneasy feeling making a user feel caused by continuing a long silence time by performing a voice response of a suitable length even when it takes a long time for reading a new recognition dictionary, and it makes the user wait. SOLUTION: Respective dictionary size information are stored beforehand in a recognition dictionary storage 4 together with plural kinds of recognition dictionaries, and when a signal processor 3 reads out newly the recognition dictionary from the storage 4, the device reads out the dictionary size information of the relevant recognition dictionary, and predicts/operates the time required for reading out the recognition dictionary based on the read out dictionary size information. Then, a response voice of an optimum length is constructed based on the predictive time required for reading out the dictionary to be outputted from a speaker 9. Thus, the situation that the long silence time continues when the signal processor 3 reads out newly the recognition dictionary is avoided, and during the reading out time, the response voice is outputted to reduce the unpleasant feeling and uneasy feeling of the user.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a voice interaction device.

[0002]

2. Description of the Related Art Conventionally, speech recognition is performed on a speech input by referring to a recognition dictionary, and another recognition dictionary is developed in response to the recognized speech input. There is known a spoken dialogue apparatus in which a user can search for necessary information only by voice input by repeating a procedure of prompting the next voice input.

[0003] In such a speech dialogue apparatus, generally, in a speech recognition process for input speech, the recognition rate decreases as the number of words to be recognized increases. In order to compensate for the drawback, the speech dictionary is hierarchized as shown in FIG. 4, and the speech input is repeated a plurality of times to advance the hierarchy, thereby narrowing down the number of words to be recognized and increasing the recognition rate. For example, to confirm the location of the Yokohama Sogo department store, take the following procedure.

[0004] Only the "address" and "facility" are registered in the recognition dictionary of the first level a. Then, when the user utters “shisetsu”, the apparatus checks the voice input for the matching degree with each of the words “address” and “facility” in the recognition dictionary, and determines the word having the highest matching degree. Is determined as a recognition word, a response voice such as "facility is output" is output, the processing is shifted to the second-level recognition dictionary e for "facility", and "facility name, e.g. Please answer ".

[0005] On the other hand, if the user speaks "department store", the degree of coincidence between each word registered in the recognition dictionary e of the second hierarchy and the input voice is calculated, and The word with the highest degree is determined as a recognized word, a response voice such as "department store" is output, and the process moves to the third-level recognition dictionary i for "department store", and then "speak the prefecture name." And a response voice is output.

In the same manner, the hierarchy of the recognition dictionary is advanced in the same manner, and when the process finally proceeds to the recognition dictionary of the fifth hierarchy k, a response voice saying "Please speak the department store name" is output. If the user utters “Sogo” and the match with the word “Sogo” in the recognition dictionary is the highest,
It is determined that this “Sogo” department store has been selected.

[0007] In this way, it is finally determined that "SOGO Department Store in Yokohama" has been inputted as a search target by voice, and the location of the corresponding facility is determined from the map information on the navigation device side. A route search with the destination as is performed, and the result is displayed on the display.

[0008]

However, in such a conventional speech dialogue apparatus, speech recognition is performed for a certain input speech, a recognition dictionary of a deeper hierarchy is read in response thereto, and speech recognition for the next speech input is performed. During the reading of the recognition dictionary, no response voice is output, or even if the response voice is output, the response time is constant and short-lived. For a long time, the user does not recognize his / her words correctly and often cannot determine whether the device is stopped or whether the device has been recognized but the dictionary has been newly read, which causes discomfort and anxiety. There was a problem that could be saved.

The present invention has been made in view of such a conventional problem, and it takes a long time to read a new recognition dictionary, so that a voice response of an appropriate length can be provided even when the user is kept waiting. It is an object of the present invention to provide a voice interactive device that can reduce discomfort and anxiety caused by a user by holding a long silent period.

[0010]

According to a first aspect of the present invention, there is provided a speech dialogue apparatus, comprising: a voice input unit; a recognition dictionary storage unit for storing a plurality of recognition dictionaries; a response voice storage unit for storing a response voice; Output means, and control means, wherein the control means calculates the degree of coincidence between the input speech from the voice input means and the recognition target word in the recognition dictionary, and based on the calculation result, the next recognition dictionary And instructing to select an appropriate response voice and output it to the voice output means, wherein the recognition dictionary storage means stores dictionary size information for each of the plurality of recognition dictionaries, Reads the dictionary size information of the recognition dictionary at the time of reading the recognition dictionary, predicts the time required to read the recognition dictionary based on the read dictionary size information, and calculates the read time. Is intended to instruct the building a response voice optimum length based on output to the audio output means estimated time required to put out.

According to a second aspect of the present invention, in the first aspect, the control means adjusts a reading speed of the response voice to construct and output the response voice having the optimum length. It instructs the audio output means.

According to a third aspect of the present invention, in the first aspect of the present invention, the control means inserts a silent portion into a predetermined position of the response voice so that the optimum length of the response voice is obtained. The voice output means is instructed to construct and output a response voice.

According to a fourth aspect of the present invention, in the first aspect, the control means selects and outputs a response voice having an optimum length from the plurality of response voices having different lengths. It instructs the audio output means.

According to a fifth aspect of the present invention, in the speech dialogue apparatus according to the first to fourth aspects, the recognition dictionary storage means is also used as a data storage means for holding data requested from another apparatus, and the control means Predicts and calculates the time required to read the recognition dictionary from the data storage means.

According to a sixth aspect of the present invention, in the first aspect, the recognition dictionary storage unit is connected to the control unit via a network, and the control unit is
The time required for reading the recognition dictionary is predicted and calculated in consideration of the load on the network.

[0016]

According to the first aspect of the present invention, a plurality of types of recognition dictionaries are stored together with the respective dictionary size information in the recognition dictionary storage means. Reads the dictionary size information of the recognition dictionary, calculates the time required for reading the recognition dictionary based on the read dictionary size information, and calculates the optimal length based on the predicted time required for the reading. It instructs a response voice to be constructed and output to the voice output means.

Thus, when the control unit reads a new recognition dictionary, it is possible to avoid a situation in which silence continues for a long time as in the prior art, and to output a response voice during the reading time. You can avoid anxiety.

According to the second aspect of the present invention, when the control unit reads a new recognition dictionary, the control unit adjusts the reading speed of the response voice to construct and output a response voice having an optimum length. By instructing the voice output means, in addition to the effect of the first aspect of the present invention, a highly continuous natural response voice can be output.

According to the third aspect of the present invention, when the control unit reads a new recognition dictionary, the control unit inserts a silent portion at a predetermined position of the response voice to thereby obtain an optimum length. By instructing the voice output means to construct and output the response voice of (1), it is possible to output a natural response voice without changing the sound quality of the synthesized voice in addition to the effect of the first aspect of the present invention.

In the voice dialogue apparatus according to the fourth aspect of the invention, the control means instructs the voice output means to select and output a response voice having an optimum length from a plurality of response voices having different lengths. In addition to the effects of the first aspect of the present invention, it is possible to output a natural sound with high continuity without changing the sound quality.

In the speech dialogue apparatus according to the fifth aspect of the present invention, the recognition dictionary storage means is also used as a data storage means for holding data requested from another apparatus, for example, a map information storage means in a navigation apparatus. 2. The control device according to claim 1, wherein the control means predicts and calculates the time required for reading the recognition dictionary from the data storage means.
In addition to the effects of the inventions of the fourth to fourth aspects, even in a device linked to another system, the time required to read a new recognition dictionary is accurately predicted, and the optimum length of the readout is determined according to the predicted time. A response voice can be output.

According to the speech dialogue apparatus of the present invention, the time required for the control means to newly read the recognition dictionary from the recognition dictionary storage means connected by the network is predicted in consideration of the load on the network. By calculating, in addition to the respective effects of the inventions of claims 1 to 4,
The time required to read a new recognition dictionary can be accurately predicted in consideration of the load on the network, and a response voice having an optimum length can be output according to the predicted time.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 shows a functional configuration of the voice interaction apparatus according to the first embodiment of the present invention. In the voice interaction apparatus shown in FIG. 1, the microphone 1 is for voice input, and the A / D converter 2 converts the voice input from the microphone 1 from analog to digital, converts it into a digital signal, and inputs the digital signal to the signal processing apparatus 3. I do. The signal processing device 3 controls the entire device and includes a CPU 3A for executing necessary arithmetic processing and an internal memory 3B having a required capacity. The external storage device 4 is shown in FIG.
And a voice recognition dictionary having a hierarchical structure as shown in FIG. 3 and response voice data, and other necessary information are stored. The switch 6 is a start / stop switch for the voice input function.

The D / A converter 7 converts the response voice data selected and edited by the signal processing
A conversion is performed. The audio signal output of the D / A converter 7 is amplified by the amplifier 8 and output from the speaker 9 as an audible sound.

Next, a description will be given of a voice response operation using the voice dialogue apparatus having the above configuration. FIG. 2 is a flowchart showing a voice response process by the voice interaction device of the present embodiment.
The external storage device 4 stores a recognition dictionary having a hierarchical structure shown in FIG. The dictionary size data S (byte) is also stored for the recognition dictionary of each hierarchy.

In the initial state of voice input, the signal processing device 3
Sets the partial speech dictionary a of the highest hierarchy of the hierarchized speech dictionary as the first recognition target (step S10). When the switch 6 is operated, the voice recognition processing is started (step S11), and the processing of steps S12 to S20 is repeated every time the dictionary hierarchy advances.

In step S12, the CPU 3A starts reading the partial speech dictionary from the external storage device 4 into the internal memory 3B. At the beginning of the dictionary reading, the dictionary size data X is read and stored in the memory 4. Since the reading of the dictionary itself is processed by parallel processing, there is no need to wait for the completion of the reading of the dictionary in step S12.
Move to

In step S13, the CPU 3A adjusts the reading speed of the response voice and outputs it. The message content of the response voice prompts entry into the dictionary, for example,
According to the hierarchy of the dictionary, "Is the intended address or facility?", "Speak the name of the state or province where you are located", "Speak the type of facility", "Department store" Please say the name of "."

In this step S13, first, the memory 3B
Of the dictionary required for reading the dictionary from the dictionary size S stored in. Set the dictionary size to S (byt
e), the time required for reading T (msec)
Is calculated using the coefficient k determined by the access speed of the external storage device 4, the CPU 3A, the internal memory 3B, and the like, and T = k × S (1). The coefficient k is stored in advance in the memory 3B for each device specification or registered at the time of setup.

Next, the CPU 3A adjusts the reading speed according to the required reading time T to construct and output a response voice having an optimum time length. For this purpose, first, the CPU 3A selects, from the memory 3B, the message content of the response voice output in step S13. If the length of this message is Tm (msec) and the normal frequency of the output D / A conversion is Fm (kHz), the frequency F (kHz) actually output is: F = Fm × (Tm / T) (2) The CPU 3A sets the frequency F to the D / A converter 7, and outputs a response voice via the D / A converter 7, the amplifier 8, and the speaker 9. As a result, the reading of the recognition dictionary is almost completed when the response voice ends (step S14).

Another method of adjusting the length of the response voice according to the time T required for reading the recognition dictionary is to insert a silent portion at a predetermined position set in advance for each response voice. is there. For this purpose, the message content of the response voice includes a special notation at a place where a silent part can be inserted, and, for example, “Enter the name of the station.
"Please speak," and insert a notation in the place where silence can be inserted. And C
The PU 3A inserts a silent part when Tm <T is satisfied, but sets the number of insertable parts of the silent part to N and the silent part 1
Assuming that the length per part is δT, a silent part having a length of δT obtained by the equation of δT = (T−Tm) / N (3) is inserted into each of the silent part insertable points △. The response voice into which the silent portion is inserted is output through the D / A converter 7, the amplifier 8, and the speaker 9.

In steps S13 and S14, a response voice is constructed and output, and a predetermined recognition dictionary is read from the external storage device 4 into the internal memory 3B. Then, in step S15, a process of capturing a user's voice input is started. Hereinafter, for the sake of simplicity, it is assumed that the response voice output first is "Please say how to set the destination".

Here, the CPU 3A operates the dialogue start switch 6
Until is pressed, the average power of the digital signal input through the A / D converter 2 is calculated. After the dialogue start switch 6 is pressed, when the instantaneous power of the digital signal becomes larger than the average power by a predetermined value or more, it is determined that the user has spoken, and the voice capturing process is started. Here, it is assumed that the user has spoken “shisetsu” in response to the response voice.

Subsequently, in step S16, the CPU 3A calculates the degree of coincidence between the partial speech dictionary read into the memory 3B and the voice section of the fetched voice input. here,
The degree of coincidence between the input speech section “shisetsu” and each of “address” and “facility” in the recognition dictionary is calculated. Note that even during the processing of this step, the voice capturing is continued by the parallel processing.

In the following step S17, when the instantaneous power of the input digital signal has become equal to or less than a predetermined value for a predetermined time or more, it is determined that the utterance of the user has ended, and the reception of the input is ended.

In step S18, the CPU 3A waits for the completion of the calculation of the matching degree, and selects the word having the highest matching degree after the completion. Here, "facility" should have a higher degree of coincidence. Therefore, the word "facility" is displayed on the monitor 5 as a result of the voice recognition (step S19).

The CPU 3A determines whether or not the recognized word indicates a partial speech dictionary below the recognition dictionary. If the word indicates a partial speech dictionary below the recognition dictionary, a new partial speech dictionary is set. Then, the process returns to step S12. Thereafter, the processing of steps S12 to S20 is repeated until the selected word does not indicate a lower partial speech dictionary (step S20).

Thus, for example, "facility"-"station"-
When the voice recognition proceeds as "Kanagawa"-"JR"-"Sakuragicho", the word "Sakuragicho" is finally output (step S21).

Thus, when one word is determined by voice dialogue, this word can be used as a search key, for example, for setting a destination in a navigation device, searching for a geographical area, and the like.

The following method can be adopted as a method for optimizing the length of the response voice according to the estimated time T required for reading the recognition dictionary. That is, a plurality of types of response voices are prepared depending on the dictionary size of the partial voice dictionary to be read. Prior to reading the partial voice dictionary to be read, the response voice is first read from the external storage device 4 and stored in the memory 3B, Alternatively, it is stored in the memory 3B in advance during setup.

For example, in the recognition dictionary shown in FIG.
When a station name is input by voice, the number of JR stations greatly differs depending on the prefecture, and therefore, the size of the JR station name dictionary also greatly differs depending on the prefecture. Therefore, as a response voice, A: "Please utter the name of the station" B: "Please utter the name of the station clearly" C: "Please utter the name of the station of JR clearly" D: Prepare different types of different lengths, such as "Please clearly state the name of the JR station in Kanagawa Prefecture", according to the size of the partial dictionary to be read, and therefore the estimated time required for reading The content with the optimal length is selected and output.

That is, the CPU 3A calculates the predicted reading time T of the recognition dictionary from the external storage device 4, and if the time T is 0 <T ≦ Tma, the CPU 3A sends the A message and Tma <T ≦ Tma.
If mb, the B message is selected, if Tmb <T ≦ Tmc, the C message is selected, and if Tmc <T, the D message is selected and output. Or in the above,
If Tmc <T ≦ Tmd, select the D message.
If Tmd <T, the output speed of the response voice is adjusted as in the first embodiment, or processing is performed so that a silent portion is inserted and output. Where Tma, T
mb, Tmc, and Tmd are values set in advance according to the device performance. In this way, it is possible to make the reading of the necessary partial recognition dictionary almost complete at the time when the output of the response voice ends.

Next, a voice interactive device according to a second embodiment of the present invention will be described with reference to FIG. The voice interactive device according to the second embodiment is characterized in that it is connected to the navigation device 10 by a network 11 and shares the external storage device 4.

The navigation device 10 detects the absolute position of the current position of the vehicle based on information from the GPS sensor or the acceleration sensor of the vehicle, and uses the map information stored in the external storage device 4 to correspond to the current position. The map information of the surrounding area is read through the network 11 and the monitor 5
To be displayed.

On the voice interactive device side, the CPU 3A accesses the external storage device 4 through the network 11, reads out the voice recognition dictionary, stores it in the internal memory 3B, and uses it for voice recognition.

Next, a speech recognition process according to the second embodiment will be described. The flowchart of the voice recognition process in the voice dialogue device of the second embodiment is the same as that of the first embodiment shown in FIG. However, the processing contents of the CPU 3A in step S12 and step S13 are as follows.

In step S12, the CPU 3A starts reading a necessary recognition dictionary from the external storage device 4 via the network 11. This reading is performed in two stages, that is, the acquisition of prior information and the reading of the recognition dictionary body.

First, in order to obtain advance information, the external storage device 4
Acquisition of the time Tn until the access is permitted, and acquisition of the recognition dictionary size information are included. The CPU 3A requests the navigation device 10 to permit access to the external storage device 4 via a network interface (not shown). The navigation device 10 receives the request,
The waiting time Tn until permission is returned. When the navigation device 10 is accessing the map information from the external storage device 4, this Tn is the time until the access ends.

The CPU 3A holds the waiting time Tn in the internal memory 3B. Only the information of the dictionary size S (byte) of the recognition dictionary to be read is read from the external storage device 4 and stored in the memory 3B.

Since the reading of the recognition dictionary itself is processed by parallel processing, there is no need to wait for the completion of the reading in step S12, and the process proceeds to the next step S13 when the advance information is obtained.

In step S13, the CPU 3A adjusts the reading speed of the response voice and outputs it. The message content of the response voice prompts entry into the dictionary, such as "speak the station name" as in the first embodiment.

On the other hand, the time T (msec) required for reading is predicted from the waiting time Tn and the dictionary size S stored in the memory 3B.

T = Tn + k × S (1 ′) where k is a constant similar to that of the first embodiment,
A different value is taken for each device depending on the access speed of the external storage device 4 via the network 11 and the like. Then, using the predicted time T, the reading speed is adjusted to construct and output a response voice having an optimal time length, as in the case of the first embodiment.

The processing after step S15 is the same as in the first embodiment. As a result, in the voice interactive device according to the second embodiment, the external storage device 4 is shared with another device, and the time required to read the recognition dictionary depends on the access state of the other device. In such an environment, the output time of the response voice can be adjusted and output according to the time required for the reading, and further, the network 1
1, it is possible to calculate the time required to read the dictionary taking into account the characteristics of the network 11 and adjust the output time of the response voice, and output the result, taking into account the characteristics of the network 11. Thus, it is possible to solve the problem that the user is given a long silent time when reading the dictionary.

In the second embodiment, as another method of adjusting the length of the response voice according to the time T required to read the recognition dictionary, a predetermined method preset for each response voice is used. A method of inserting a silent part at a position can be adopted.

Further, a plurality of types of response voices are prepared according to the dictionary size of the partial voice dictionary to be read, and the response voice is first read from the external storage device 4 before reading the partial voice dictionary to be read, and is read from the memory 3B. In the memory or in the memory 3B in advance during setup.
And a method of selecting and outputting a content having an optimal length according to the size of the partial dictionary to be read, that is, the estimated time required for reading, can also be adopted.

As described above, according to the speech dialogue apparatus of the present invention, when a new recognition dictionary is read, the dictionary size information of the recognition dictionary is read, and the recognition dictionary is read based on the read dictionary size information. The CPU 3A constructs and outputs a response voice having an optimal length based on the prediction time required for the readout, and based on the prediction time required for the readout.
When newly reading the recognition dictionary from the external storage device 4,
It is possible to avoid a situation in which silence continues for a long time as in the related art, and to output a response voice during the readout time, thereby preventing the user from feeling uncomfortable or uneasy.

When reading the recognition dictionary from the external storage device 4, the CPU 3A adjusts the reading speed of the response voice to output a response voice having an optimum length, thereby providing a natural response with high continuity. Sound can be output.

When reading the recognition dictionary from the external storage device 4, the CPU 3A inserts as many silent portions as necessary into the predetermined position of the response voice, thereby constructing a response voice having an optimum length. Thus, a natural response voice can be output without changing the sound quality of the synthesized voice.

Further, when reading the recognition dictionary from the external storage device 4, the CPU 3A selects and outputs a response voice having a length corresponding to the size of the recognition dictionary, thereby changing the sound quality without changing the sound quality. It is possible to output a natural response voice with high performance.

Further, in the voice dialogue apparatus of the present invention, when the external storage device 4 is also used as data storage means for holding data requested from another device 10, the CPU 3
A predicts and calculates the time required for reading the recognition dictionary from such an external storage device 4 so that the time required for newly reading the recognition dictionary can be accurately determined even in a device linked with another device. And a response voice having an optimal length can be output according to the prediction time.

In addition, in the spoken dialogue apparatus of the present invention, the time required for newly reading out the recognition dictionary from the external storage device 4 connected to the network 11 is calculated by taking the load of the network 11 into consideration. By doing so, it is possible to accurately predict the time required to read a new recognition dictionary, and to output a response voice having an optimum length according to the predicted time.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention.

FIG. 2 is a flowchart of a speech recognition process according to the embodiment.

FIG. 3 is a block diagram showing a configuration according to a second embodiment of the present invention.

FIG. 4 is a structural diagram of a speech recognition dictionary having a general hierarchical structure.

[Explanation of symbols]

 Reference Signs List 1 microphone 2 A / D converter 3 signal processing device 3A CPU 3B memory 4 external storage device 5 monitor 6 switch 7 D / A converter 8 amplifier 9 speaker 10 navigation device 11 network

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 3/00 571U

Claims (6)

[Claims] 1. A speech input means, a recognition dictionary storage means for storing a plurality of recognition dictionaries, a response speech storage means for storing a response speech, a speech output means, and a control means, wherein the control means comprises: Calculating the degree of coincidence between the input voice from the voice input means and the recognition target word in the recognition dictionary, reading the next recognition dictionary based on the calculation result, and selecting an appropriate response voice to output the voice In the voice interactive device for instructing the recognition dictionary to output, the recognition dictionary storage unit stores dictionary size information for each of the plurality of recognition dictionaries, and the control unit reads the recognition dictionary when reading the recognition dictionary. The dictionary size information is read, the time required for reading the recognition dictionary is calculated based on the read dictionary size information, and the optimum length is calculated based on the estimated time required for reading. Speech dialogue system to build a response voice, characterized in that an instruction to output to the sound output unit. 2. The apparatus according to claim 1, wherein the control unit instructs the voice output unit to adjust a reading speed of the response voice to construct and output the response voice having the optimum length. The speech dialogue device according to 1. 3. The sound output means to construct and output a response sound of the optimal length by inserting a silent part at a predetermined position of the response sound, the control means being configured to output the response sound. The voice interaction device according to claim 1, wherein the instruction is given. 4. The apparatus according to claim 1, wherein the control unit instructs the voice output unit to select and output a response voice having an optimum length from among the plurality of response voices having different lengths. 2. The voice interaction device according to 1. 5. The recognition dictionary storage means is also used as a data storage means for holding data required from another device, and the control means is configured to read a time required for reading the recognition dictionary from the data storage means. The speech dialogue device according to any one of claims 1 to 4, wherein the speech dialogue device performs a prediction operation. 6. The recognition dictionary storage means is connected to the control means by a network, and the control means predicts and calculates a time required for reading the recognition dictionary in consideration of a load on the network. The voice interaction device according to claim 1, wherein