JPH04167749A - Audio response equipment - Google Patents

Abstract

PURPOSE:To output an audio message with high quality easily listened to as a response by obtaining an audio by means of the rule synthesis system from only a part of a response message from which the audio is not obtained by a recording edit system. CONSTITUTION:This equipment consists of a telephone set 1, a communication line 2, a line control section 3, a PB signal reception transmission section 4, an audio recognition section 5, a host computer 6, a customer code conversion table 7, a response section (1st audio response section by recording edit system) 8, an audio rule synthesis section (2nd audio response section by rule synthesis system) 9 and an accept dictionary 10. When the information with respect to the recognition result of the entered information is outputted in audio as a response, the audio outputted basically as a response is generated by the recording edit system and the audio not generated by the recording edit system is generated and outputted by the rule synthesis system. Thus, the audio outputted as a response is brought into the audio with high quality easily listened to.

Description

DETAILED DESCRIPTION OF THE INVENTION [Purpose of the Invention (Field of Industrial Application) The present invention relates to a voice response device capable of outputting response information to input information in an easily audible voice.

(Prior art) Recently, in telephone services in banking, etc., and event information services in department stores, etc., inquiry information input via the telephone as a voice or a phone signal is recognized, and response information in response to the recognition result is provided by voice. Voice response devices that provide notifications are attracting attention.

With this type of voice response device, for example, the response content is predetermined.Thank you for using the device. ”, “We would like to inform you about your inquiry.”, and non-standard sentences that include response information that changes in various ways depending on the content of the inquiry.The above-mentioned standard sentences are generated using a recording editing method, and various types of sentences are generated. Attempts have been made to generate □ non-standard sentences that change in a variety of ways using a rule composition method.

Note that the recording editing method combines and edits various voice messages that have been recorded and collected in advance to obtain the voice to be output as a response, and by introducing technologies such as PCM and ADPCM, it is possible to obtain high-quality response voices. It has the characteristic that it can be done. However, there is a limit to the number of voice messages that can be recorded in advance and stored as voice files. Therefore,
It is difficult to generate voice by dealing with all of the response information that changes in a wide variety of ways. Further, it is difficult to additionally register voice data, and the task of editing various voice data in accordance with the fixed phrase to be output as a response is also extremely difficult.

In this respect, the rule composition method uses parts of speech for words.

It is equipped with an accent dictionary that stores information such as pronunciations and accent types, and uses response information given as character strings to synthesize speech in a regular manner by referring to the accent dictionary. However, it is relatively easy to generate accented speech.

However, with this recording/editing method, if a character string (non-standard sentence) that is not registered in the accent dictionary is given, it is not possible to add an accent to the voice that is synthesized according to the rules. It is only possible to generate flat, unaccented voices. Moreover, when compared with the audio generated by the recording/editing method, the quality of the audio is undeniably inferior, and there is also the problem that it is difficult for those who are not familiar with the rule-synthesized audio to hear it.

(Problem to be solved by the invention) In this way, in the past, the information content to be output as a response is divided into fixed sentences and non-fixed sentences, and audio is generated for the fixed sentences using a recording/editing method. As for sentences, speech is generated using rule synthesis, so only fixed sentences that are not very important for the response content are output in high quality, while important content that should be output as a response is generated by rule synthesis and is difficult to hear. There was a problem with the output being output as audio.

The present invention has been made in consideration of the above circumstances, and its purpose is to provide a highly practical voice response device that can output a voice of high quality that is easy to hear. There is a particular thing.

[Structure of the Invention] (Means for Solving the Problems) When the voice response device according to the present invention recognizes input information and outputs information in response to the recognition result by voice, the voice response device basically responds by The audio to be output is generated using the recording/editing method, and the audio that cannot be generated using this recording/editing method is
In other words, when the voice to be generated and output is not registered in the voice response section of the recording/editing method, the voice is generated and output by the rule synthesis method.

In other words, instead of dividing the information content to be output in response into fixed and non-fixed sentences, even if it is a non-fixed text, the audio data of frequently used sentences is registered and the recording/editing method 1 is used. The system is characterized in that a voice is generated and only those voices that are not registered in the voice response section of this recording/editing method are generated and output using a rule synthesis method.

(Function) According to the present invention configured as described above, it is possible to output responses to frequently used non-standard sentences, such as names of major customers, using high-quality audio using a recording and editing method. Only when a non-standard sentence consisting of unregistered special information content is to be output as a response, the speech will be generated and output using the rule synthesis method. As a result, although some low quality audio generated by rule synthesis is output,
Overall, it becomes possible to output a large amount of high-quality voice generated by the recording/editing method, and it becomes possible to improve the quality of the overall response output voice and make it easier to hear.

(Embodiment) Hereinafter, a voice response device according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of the embodiment device, where ◯ indicates a telephone connected to the embodiment device via the communication line 2. In FIG. A user of this device inputs inquiry information via the telephone 1 using voice or a PB double signal (buzz phone signal), and receives the service by receiving the response information as a voice message.

Note that if a dedicated terminal device is used instead of the telephone 1, inquiry information may be input as the code information.

Therefore, the line control unit 3 controls the line connection with the telephone 1 connected via the communication line 2, and
Send and receive audio, PB double signal code information, etc.

The PB signal reception/transmission unit 4 identifies the PB signal input from the communication line 2 via the line control unit 3, provides the identification code information to the host computer 6, and also identifies the PB signal for automatic calling of the telephone 1. A double signal is generated under the control of the host computer 6, and sent from the line control section 3 to the communication line 2 for automatic calling. The voice recognition unit 5 recognizes the voice input from the communication line 2 via the line control unit 3 and provides the recognition result to the host computer 6.

Inquiry information input from the telephone 1 using voice or PB multiplied signals is recognized and identified by the voice recognition section 5 and PB signal reception/transmission section 4, and then given to the host computer 6 to read the information content. is interpreted. Then, in the host computer 6, according to the interpretation result (inquiry information),
For example, by searching a database not shown,
Requests answer information consisting of character strings, etc. Then, depending on the answer information, a response unit 8, which is a first voice response unit, generates a voice message indicating the above response information by a recording/editing method, or a second voice response unit, which generates a voice message indicating the above response information by a rule synthesis method. The voice rule synthesis unit 9, which is the voice response unit No. 2, is activated to generate the voice to be output as a response. These responses are generated by the response unit 8 and the speech rule synthesis unit 9.
The voice is output to the communication line 2 via the line control section 3, and is output as a response to the telephone set 1.

Note that the speech rule synthesis unit 9 refers to an accent dictionary 10 that stores information contents as shown in FIG. 2, for example, and generates the speech indicated by the input character string by adding accents and cents as appropriate. It is something. This rule-based synthesis of speech is performed by appropriately utilizing various algorithms for rule-based synthesis of speech that have been proposed in the past.

Now, the feature of this embodiment device is that it is equipped with a customer code conversion table 7 configured as shown in FIG. It has a function of referring to table 7 and checking whether the voice data of the answer information to which the voice response is to be made is registered in the response section 8 or not.

Then, in the host computer 6, if the voice data of the answer information to which a voice response is to be made is registered in the response section 8,
The response unit 8 is activated to generate a voice using the recording/editing method and output as a response, and only when the voice rule synthesis unit 10 is not registered, the voice rule synthesis unit 10 is activated to synthesize voice using the rule synthesis method. It is characterized by

That is, in the customer code conversion table 7, for example, customer names are stored in order of frequency of use in association with phrase numbers of voice data registered in the response section 8. When the host computer 6 is asked for answer information,
The customer code conversion table 7 is searched for each word or phrase constituting the character string representing the answer information, and it is checked whether the word or phrase is registered. If a corresponding word or phrase (in this example, customer name) is registered in the customer code conversion table, the phrase number associated with the word or phrase is obtained and output to the response section 8. The corresponding audio data is read out, and the audio is generated and output.

Further, if the corresponding word or phrase is not registered in the customer code conversion table 7, the host computer 6 provides the character string indicating the word or phrase to the speech rule synthesis section 9, and refers to the accent dictionary 10. This will cause the speech to be synthesized according to rules.

Specifically, this storage configured as described above operates as follows.

When there is a call from the telephone 1, the line control unit 3 detects this and starts the host computer 6, for example, the code number 1.
An initial operation is performed to inquire about the account number and transfer introduction. In response to this, host computer B responds by outputting a series of messages such as ``This is O○ Bank.'' ``Please enter your service code.'' ``Please enter your account number and PIN number.'' Phrase numbers specifying these response contents are given to the response unit 8.

In response to this, the response unit 8 reads data corresponding to the voice message specified by the phrase number from the response memory, converts it into a voice waveform, generates a voice signal, and responds to the telephone 1. Output.

In accordance with this response message, the user of the telephone 1 inputs the PB double signal by, for example, a dial bat operation, and inputs the account number or password number prompted by the message. Needless to say, this information may be manually inputted by voice.

Then, in this device, the PB signal reception and transmission section 4 identifies the input PB multiplied signal, and the identification information is transmitted to the host computer 6.
give to If the input is a voice, the input voice is recognized by the voice recognition section 5, and the recognition result (for example, a numeric code) is provided to the host computer 6. By analyzing the information input and processed in this way, the host computer 6 can determine whether the user of the telephone 1 is requesting information (
It will then know whether the user is making an inquiry) and will request response information in response to the request.

This response information is obtained as, for example, "3 transfers", "There was a transfer", etc. This is the underlined part or atypical part of this response message.

The host computer 6 sequentially refers to the customer code conversion table 7 for the words indicated by these character strings to check whether the voice data is registered or not, and according to the results, the synthesis unit 8 converts the words to the customer code conversion table 7. It is determined whether to generate the voice or to have the voice rule synthesis unit 9 generate the voice. Here, we will explain the underlined character string (customer name), which is the non-standard part. The host computer 6 can output the underlined character string part (customer name) as a voice using the Recording East method. Check whether it exists or not.

In this case, since "Kawazaki Kikyo" is not registered in the customer coat conversion table 7, the character string information is given to the voice rule synthesis unit 9, and the above character string is generated by the rule synthesis method referring to the accent dictionary 10. Synthesize and output the voice that indicates.

Mata "Yokohama Kogyo" is registered in the customer code conversion table 7, so the phrase number [
201] is given to the response section 8, which synthesizes and outputs the voice representing the above character string using the recording/editing method. Since the next "Kawasaki Kogyo" is not registered in the customer code conversion table 7, the character string information is given to the voice rule synthesis section 9, and the above character string is similarly synthesized by the rule synthesis method referring to the accent dictionary 10. Synthesize and output the audio indicated by the column.

In addition, in this rule composition, as shown in Figure 2, the accent type for "Kawasaki" is specified as "0", and the accent type for "Kigyou" and "Kougyo" is specified as [1], so Using [′] as the accent position, the speech is synthesized in a regular manner, such as ``Kawasaki kigyo'' and ``Kawasaki kogyo''.

As described above, according to the device of this embodiment, the audio data indicating the audio data (customer name) of the non-standard part registered due to high frequency of use, etc. in the voice message to be output as a response is recorded and edited. Since it is generated and output, it is possible to create a highly natural and high-quality voice message.

Then, the voice data (customer name) of the unregistered atypical part is tracked and the voice is generated and output using the rule synthesis method, which increases the overall amount of voice generated using the recording and editing method, making it high quality and easy to hear. can be taken as a thing. As a result, great practical effects can be achieved, such as making it possible to efficiently respond to voice messages with less discomfort.

Note that the present invention is not limited to the embodiments described above. For example, the configuration of the accent dictionary 10 and the information on customer names registered in the table 7 are not limited to the examples described above. In other words, it is of course possible to limit the voice generated by the recording/editing method to the customer's name or to use other frequently used words or phrases. Alternatively, the accent dictionary 10 and the character string may be checked in the host computer, and the phonetic rule synthesis section 9 may be provided with the prosody information and the phoneme sequence to execute the rule synthesis process. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Effects of the Invention] As explained above, according to the present invention, instead of dividing a voice message into regular sentences and non-standard sentences and generating voices separately for each, it is not possible to obtain the voices by using a recording editing method. Since the voice of only the response message is obtained using the rule synthesis method, by registering voice data for frequently used words and phrases in advance, it is possible to increase the overall voice generated by the recording/editing method. However, great practical effects can be achieved, such as making it possible to output a high-quality voice message that is easy to hear.

[Brief explanation of the drawing]

The figures show a voice response device according to an embodiment of the present invention, and FIG. ,
FIG. 3 is a diagram showing the contents of the customer code conversion table in the embodiment. 3... Line control unit, 4... PB signal reception and transmission unit, 5
. . . speech recognition unit, 6 . . . host computer, 7 . . . customer code conversion table, deceased . second voice response unit based on rule synthesis method), 10...accent dictionary. Applicant's agent Patent attorney Takehiko Suzue

Claims (1)

[Scope of Claims] A voice response device that recognizes input information and outputs information in response to the recognition result as a voice response, comprising: a first voice response unit that generates voice to be output in response using a recording and editing method; A voice response device comprising: a second voice response unit that generates a voice to be output as a response by a rule synthesis method when the voice to be outputted is not registered in the voice response unit.