JPH10124086A - Voice interactive system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a system which can obtain the natural interaction between an unspecified speaker and a machine. SOLUTION: A user's voice is received by a voice input device, a communication system 12 and a voice reception data 13, and an active noise filter 14 removes noise; and a voice recognition device 15 which uses unspecified-speaker word voice recognition technology performs speech recognition and a voice understanding device 16 extracts meaning contents from the voice. Then an expert system 17 having a knowledge base of information that users request and limited information obtains answer voice information for the users and when there is not the information, a person intervenes with a voice through a system supporter 18 to send speech information from a voice output device 21 to the users with a voice or voice signal synthesized by a voice synthesizing device 19.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech dialogue system in which an unspecified speaker and a computer interact with each other by voice.

[0002]

2. Description of the Related Art Hitherto, in a dialogue between a person and a machine, when a user listens to various information recorded in advance on a tape recorder or the like by telephone or collects information using a facsimile, a person follows an announcement. You are operating a machine. In this case, the voice information that can be generated by the machine side is limited, and since a person responds by operating the machine, the content of the dialogue and the response time are greatly limited.

In order to eliminate these restrictions,
2. Description of the Related Art There is a speech synthesizer based on a rule synthesis method in which speech information that can be generated by a machine is arbitrary. Also, people respond by voice,
There is a voice recognition / understanding system that recognizes this on the machine side, and a voice interface for multimedia that handles image information and voice information.

FIG. 7 is a block diagram showing a speech synthesis process of the Japanese rule synthesis system. The Japanese parsing unit 1 analyzes reading / word / phrase boundaries and the like by syntactic analysis and semantic analysis with reference to the Japanese dictionary 2 for kanji / kana mixed sentences input from a personal computer, word processor, workstation or the like. . From this analysis result, the accent and intonation of phoneme series and the duration control unit 4 by the intonation control unit 3
, The parameters of the duration of the phoneme are determined, and these parameters are used as parameters of the acoustic tube model to perform voice synthesis with reference to the voice dictionary 6 by the voice synthesis unit 5, thereby obtaining a synthesized voice from the loudspeaker 7.

In order to use such a speech synthesizer in a dialogue system, speech synthesis is performed by a device having a line control mechanism such as a connection response from outside the line, a connection request to the outside of the line, and a push button signal recognition function of a telephone. A device is provided, and a voice signal is directly output from a modular terminal of the voice synthesizing unit 5 to a telephone line. With this device, a telephone-compatible system can be constructed simply by connecting it to a computer.
Simultaneous response and simultaneous notification of up to two lines are possible.

Next, the speech recognition apparatus performs word speech recognition, continuous word speech recognition, and speech understanding by performing acoustic analysis, phoneme recognition, word recognition, and language processing of the input speech. For example, as shown in FIG. 8, the word extraction type speech recognition system roughly extracts necessary words by fuzzy inference, registers necessary words in advance, and gradually adds them later by a learning function. Since it is a system to go and can be operated by software, it can be used as a voice recognition means in a dialogue between a computer and a person.

As a voice interface for multimedia, as shown in FIG. 9, as a part of realizing multimedia technology using a moving image processing unit and voice response, VME
There is one in which a multitasking-compatible voice input interface is constructed on the X window of a workstation using a high-precision unspecified speaker word voice recognition technology developed on a bus-compatible high-speed neurocomputer.

The features of this system include (a) voice input to a focus window, (b) voice dictionary switching corresponding to a task, and (c) operation equivalent to a mouse for an application. can give.

[0009]

As means for dialogue between humans and machines, there have conventionally been voice synthesizers, voice recognition / understanding systems, and voice interfaces for multimedia, but these are simply combined to create a voice dialogue system. By constructing, it is impossible to realize a human interface as if an unspecified person has a natural conversation.

It is an object of the present invention to provide a system capable of obtaining a natural conversation between an unspecified speaker and a machine.

[0011]

SUMMARY OF THE INVENTION The present invention provides a high-speed neuron for a VME bus as a technology capable of coping with an unspecified speaker, in order to realize a human interface as if a person has a natural conversation. It has a knowledge base and a learning function that uses the high-precision unspecified speaker word speech recognition technology developed on the computer and can be handled by rules in order to extend the information and limited information required by the user. It integrates an expert system and a speech synthesis system that responds to user requests and obtains a more natural and near-human voice output, and implements a human interface that can operate on a workstation. An audio signal receiving means for receiving an audio of a user via a communication system and obtaining an audio signal from which noise has been removed; A / D conversion data of the voice signal is roughly analyzed by a frequency spectrum, and a time-series feature parameter of the voice signal is extracted for each phoneme unit by feature extraction, and word recognition and syntax recognition following recognition of each phoneme unit are performed. A voice recognition device for obtaining character data corresponding to a voice signal by semantic recognition, a voice understanding device for extracting the meaning of the character data, and a knowledge base when the voice understanding device fails to understand the meaning of the character data. An expert system that obtains speech information of speech incomprehension online using inference using, and obtains speech information online to answer user requests by inference using knowledge base when the meaning can be understood. If the voice information requested by the user is not in the knowledge base or cannot be handled, a system A voice synthesizer that synthesizes voice information from the expert system as a voice signal, and a voice signal transmitting unit that transmits a voice signal from the voice synthesizer or a voice signal from a system supporter to a user via a communication system. , Is provided.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a system block diagram showing an embodiment of the present invention. FIGS. Hereinafter, the configuration and processing of each unit will be described in detail.

The voice input device 11 is for a user to input voice from a microphone of a receiver such as a telephone. The communication system 12 performs communication control using, for example, a 2B + D B-ISDN having a high speed and a wide band with a digital line and an ATM switching system for controlling a network, and transmits a voice signal from the voice input device 11 to the voice receiving device 13. I do. Active noise filter 1
4 removes noise from the audio signal received by the audio receiving device 13.

The speech recognition device 15 is a highly accurate unspecified speaker word speech recognition technology developed on a VME bus-compatible high-speed neurocomputer as a technology capable of coping with unspecified speakers in order to recognize and understand the received speech. consists of a neural computer system includes a speech recognition processing unit 15 ₁ and the voice analysis learning unit 15 ₂ and the voice dictionary 15 _3.

As shown in FIG. 2 and FIG. 3 in the process of using the speech recognition library, the apparatus 15 performs speech understanding processing and A / D conversion processing on the input signal, and then performs speech analysis on the speech input signal. Is roughly analyzed by a frequency spectrum, feature parameters of a speech signal are extracted and converted in time series from speech analysis results by feature extraction, and segmentation into speech units is performed by a segmentation process.

Next, a phoneme sequence is obtained by performing phoneme recognition by comparison with a phoneme standard pattern by phoneme recognition, and a word is recognized by collating a phoneme sequence with a word standard pattern in a knowledge base by word matching and word recognition. . If there is no standard pattern in the recognition, a learning process (recognition / registration) is performed to add word knowledge to the knowledge base.

Next, the recognized words are analyzed for syntactic errors by comparing them with the syntax patterns of the knowledge base by syntactic matching and syntactic recognition, and if there are errors, re-verification is performed to perform syntactic recognition. Do. Furthermore, it checks whether the word syntax recognized by the semantic analysis and the semantic recognition is semantically valid, and repeats the syntactic analysis processing and the semantic analysis processing until a valid result is obtained. In these analyses, characters that can be converted by character conversion processing are converted into kanji and kana, including whether or not they can be converted into characters.

Returning to FIG. 1, for the information recognized by the speech recognition device 15 and having undergone the character conversion, the meaning understanding contained in the speech is extracted by the speech understanding device 16.

The expert system 17 is different from a conventional offline knowledge base system in that real-time inference is performed while exchanging data with another process on the same computer or another computer connected via a communication line in an online environment. With the ability to do

[0020] The expert system 17 comprises a first knowledge base 17 ₁ and the second knowledge base 17 ₂ made in the knowledge base library, as shown in FIG. 3, speech understanding by the speech understanding device 16 has failed sometimes performed inference using the knowledge base 17 ₂ of the rules, obtaining speech information of the content to notify the user that a voice misunderstanding with this knowledge base 17 _2. The expert system 1
7, gradually increasing the storage information for the knowledge base 17 ₂ has a learning function.

When the speech understanding device 16 can understand the speech content, the expert system 17
Inference using the rules of 7 ₁ is performed, and this knowledge base 17 ₁
Obtain various audio information to answer the user's request.

The system supporter 18 is used by a person for mediating interactive processing when voice information requested by the user is not available in the knowledge base or cannot be handled.

The voice synthesizer 19 synthesizes voice information from the expert system 17 as a voice signal. This device 19 is equivalent to a conventional speech synthesizer,
Obtain a more natural and close to human voice output.

The communication system 20, as shown in FIG.
The voice signal synthesized by the voice synthesizer 19 or the raw voice from the system supporter 18 is converted to B-ISDN and ATM.
The voice is transmitted to the user by an exchange system or the like, and the voice is reproduced by the voice output device 21.

The interface required for the above system block configuration is as shown in FIG. The system configuration is as shown in FIG.

With the above configuration, in this embodiment,
A human interface is realized as if a person has a natural conversation between a user and a machine.

[0027]

As described above, according to the present invention, an unspecified speaker word speech recognition technology capable of coping with an unspecified speaker is used.
And an expert system with a knowledge base and learning function that can be handled by rules to extend the information requested or limited by the user, and more natural and human voice to respond to the user's request Since a voice synthesis system that obtains a close voice output is integrated, it is possible to realize a human interface that makes it possible for an unspecified person to have a natural conversation in a human-machine conversation.

[Brief description of the drawings]

FIG. 1 is a block diagram of a voice dialogue system according to an embodiment of the present invention.

FIG. 2 is a processing step (part 1) in FIG. 1;

FIG. 3 is a processing step (part 2) in FIG. 1;

FIG. 4 is a process (part 3) in FIG. 1;

FIG. 5 is an example of an interface configuration according to the embodiment;

FIG. 6 is a system configuration example according to the embodiment;

FIG. 7 is a block diagram of a unit-type speech synthesis process.

FIG. 8 is a spectrum and an example of extraction of word speech recognition.

FIG. 9 is an example of a voice recognition interface.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Voice input device 12, 20 ... Communication system 13 ... Voice receiving device 14 ... Active noise filter 15 ... Voice recognition device 16 ... Voice understanding device 17 ... Expert system 18 ... System supporter 19 ... Voice synthesis device 21 ... Voice output device

Claims (1)

[Claims] 1. An audio signal receiving means for receiving a user's voice through a communication system and obtaining a noise-free audio signal; and roughly analyzing A / D conversion data of the audio signal by a frequency spectrum; A voice recognition device that extracts a time-series feature parameter of a voice signal in phoneme units by feature extraction, obtains word data following recognition of each phoneme unit, and character data corresponding to the voice signal by syntax recognition and semantic recognition, A voice comprehension device for extracting the meaning content of the character data, and when the voice comprehension device fails to understand the meaning content of the character data, obtain speech information of voice incomprehension online by inference using a knowledge base, An expert system that obtains voice information online to answer user requests by inference using the knowledge base when the content can be understood. A system supporter for performing a human-mediated dialogue process when voice information to be performed is not present in the knowledge base or when the voice information cannot be handled; a voice synthesizer that synthesizes voice information from the expert system as a voice signal; A voice signal transmitting means for transmitting a voice signal from the synthesizer or a voice signal from a system supporter to a user via a communication system.