JP2004012993A - Network voice recognition system, method of recognizing network voice, and network voice recognition program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To realize an economically usable system by constantly utilizing the latest voice recognition technique by maintaining a high recognition rate so that a plurality of electric apparatuses connected to a network may be operated by a human friendly operation system being vocie recognition. <P>SOLUTION: The network voice recognition system has the plurality of electric apparatuses 4, 5 having a means 1 for converting a voice into an electrical signal, and a central processing means 9 connected to the electric apparatuses through a line. The central processing means 9 executes voice recognition processing for a transmitted voice signal and transmits specified operation output to the electric apparatuses based on a processing result. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a network technology for operating a plurality of electric devices by voice recognition, and performs centralized voice recognition processing of voices of an operator instructed by a plurality of devices connected to a line by a central processing unit. The present invention relates to a network voice recognition system, a network voice recognition method, and a network voice recognition program for transmitting a predetermined operation output to an electric device.
[0002]
[Prior art]
Japanese Patent Laid-Open Publication No. Sho 56-88503 discloses a technique for handling speech recognition. In this method, a heating device is provided with a voice recognition unit and a voice synthesis unit, and by sequentially switching between two voice recognition modes and one voice synthesis mode, the operation proceeds as if the operator and the device interact with each other. This is intended to prevent malfunction due to erroneous recognition.
[0003]
The features of the input voice are extracted and stored in the memory. The recognition process compares a standard pattern stored in the ROM in advance with the pattern stored in the memory, and recognizes a pattern having a high degree of similarity as the voice command. Next, the mode is switched to the voice synthesis mode, and a predetermined voice message is issued to the operator based on the voice recognized by the voice synthesis synthesizer. When the operator listens to this message and inputs the next voice, the device shifts again to the second voice recognition mode, and determines whether a predetermined voice is input.
[0004]
With such a configuration, the system can reliably advance the control. Malfunctions such as starting heating not instructed by erroneous recognition or interrupting heating that has not yet been completed can be prevented beforehand.
[0005]
This is an application filed by the inventor of the present invention. At that time, it is reminiscent of the technical background at the time when it was difficult to perform perfect speaker recognition even with specific speaker speech recognition (a method of registering the language of the person to be recognized in advance). It is an invention.
[0006]
In recent years, with the remarkable progress of semiconductor technology and the development of software technology and network technology, information devices such as personal computers and mobile phones have become widespread in homes. Also, electrical appliances called Internet home appliances are gradually appearing on the market.
[0007]
Under such circumstances, the speech recognition technology, which has been attracting attention for a long time, is finally entering the practical stage as a technology capable of achieving good communication between humans and machines. In particular, memory has become inexpensive, CPU processing speed has increased dramatically, and speech recognition technology has evolved, making speaker-independent speech recognition very difficult ten years ago (no need for language registration in advance). (A method targeting an unspecified majority) with a high recognition rate. Computers with microphones instead of keyboards are now at a practical level. PCs with such voice recognition software pre-installed at the factory are no longer rare.
[0008]
However, a heating device as disclosed in JP-A-56-88503 has not yet been commercialized. Despite the rapid spread of speech recognition technology in the PC world. This comes solely from the economics of speech recognition systems. A PC with a fast processing speed and a large memory on a personal computer itself will enhance the value of the personal computer itself. On a personal computer, voice recognition can be introduced without increasing the price. In addition, even if a part of the information is erroneously recognized, it may be input again. It is equivalent to simply mistouching the keyboard. However, mis-recognition can lead to excessive heating and combustion in the cooking cabinet. If a speech recognition system is incorporated in the heating device, the recognition accuracy is even more important, and any attempt to increase it will add to the system cost.
[0009]
Japanese Patent Application Laid-Open No. 2000-111057 is shown as a related art of the present invention. Here, a data processing device for kitchen appliances is disclosed. A kitchen appliance, for example, a microwave oven, is provided with a microphone that generates digital data in response to an operator's voice, and voice recognition means for extracting a digital signal from the microphone. The input to the microphone is an item of food consumed in the refrigerator, and the main body is provided with storage means for storing digital signals for generating a list of processed items. This digital data is also transferred to a remote management system.
[0010]
With this arrangement, when a food item consumed in the refrigerator is input into the microphone by voice, voice recognition software loaded on the appliance body identifies this and generates an e-mail message that is forwarded to the Internet. used. The message transmitted to the Internet is used for replenishment and delivery of items.
[0011]
Also in this configuration, the kitchen appliance is provided with the voice recognition means, and it cannot be denied that the cost of the kitchen appliance is increased. Further, only information of food item names is handled by voice recognition. Even if misrecognition is performed, ordering of supplementary items is not performed accurately. It does not include a function for operating / stopping the device main body disclosed in Japanese Patent Application Laid-Open No. 2000-111057.
[0012]
[Problems to be solved by the invention]
In the present invention, a plurality of electrical devices connected to the network can be operated by a human-friendly operation system called voice recognition, so that the latest voice recognition technology is always used, and a high recognition rate is maintained while economically. Aim to realize a usable system.
[0013]
[Means for Solving the Problems]
In order to solve the above-mentioned conventional problems, a network voice recognition system according to the present invention includes a plurality of electrical devices having individual IDs provided with means for converting voice into electrical signals, and is connected to the device group by a line. And a central processing unit that performs a voice recognition process on the transmitted voice signal and transmits a predetermined operation output to the electric device based on the processing result. As a result, costly voice recognition means for realizing a high recognition rate can be placed in the central processing means, and the voice data transmitted from the electric equipment connected via the line is recognized one after another, and the operation based on the recognition result is performed. A signal can be returned.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
According to the first aspect of the present invention, there is provided an electric device having means for converting a sound into an electric signal, and a central processing means connected to the device by a line, wherein the central processing means transmits the transmitted sound signal. And a predetermined operation output is transmitted to the electric device based on the processing result. This allows the customer to economically realize easy operation by voice recognition.
[0015]
According to a second aspect of the present invention, there is provided the network voice recognition system according to the first aspect, further comprising a plurality of electric devices. Each of the devices does not need to have a voice recognition unit, and is more economical. Easy operation by recognition is available. In addition, all devices can be operated by the same operation system.
[0016]
According to a third aspect of the present invention, in the network voice recognition system according to the first aspect, the electrical devices are configured to have individual IDs, and no matter how many electrical devices are connected to the central processing unit. The central processing means can steadily identify individual devices and perform appropriate recognition. Further, if the device name, the maker / model name, and the like can be identified based on the ID, the central processing unit can grasp a group of devices connected to a certain home network, and can recognize a certain voice. An appropriate operation output can be transmitted to a plurality of devices in response to an input.
[0017]
According to a fourth aspect of the present invention, in the network voice recognition system of the first aspect, the electric device further includes an A / D converter, converts the voice into a digital signal, and transmits the digital signal to the central processing unit via a line. With such a configuration, it is easy to perform multiplex communication resistant to noise, and it is easy to shorten the communication time by compression or the like.
[0018]
According to a fifth aspect of the present invention, in the network speech recognition system of the first aspect, the speech recognition processing performed by the central processing unit includes linear prediction analysis (LPC), vector quantization (VQ), and a hidden Markov model ( HMM) based speech recognition processing, feature extraction is efficiently performed by linear prediction analysis (LPC), and the obtained feature vector can be converted into a finite number of symbols by vector quantization (VQ). Using this, a hidden Markov model (HMM), that is, a phoneme probability model (probability automaton) is created, and phoneme unit recognition can be performed.
[0019]
According to a sixth aspect of the present invention, in the network speech recognition system according to the first aspect, when the central processing unit needs to transmit a message to an electric device based on a result of the speech recognition processing, a predetermined speech is stored in the memory. Are synthesized and transmitted via a line, and the device control can be advanced while interacting.
[0020]
According to a seventh aspect of the present invention, in the network speech recognition system of the first aspect, the central processing means has a hardware for performing voice recognition and a program for performing software processing. An economical system can be realized by distributing the voice recognition processing to hardware and software in a well-balanced manner. In particular, the load on the central processing unit can be reduced.
[0021]
According to an eighth aspect of the present invention, the audio electric device converts sound into an electric signal, the electric device transmits an audio signal, and the central processing means performs a voice recognition process on the transmitted audio signal. The network recognition method includes a step and a step of transmitting a predetermined operation output to the electric device based on a processing result. This allows the customer to economically realize easy operation by voice recognition.
[0022]
According to a ninth aspect of the present invention, the electric device converts voice into an electric signal, the electric device transmits a voice signal, and the central processing means performs a voice recognition process on the transmitted voice signal. And a step of transmitting a predetermined operation output to the electric device based on the processing result. This allows the customer to economically realize easy operation by voice recognition.
[0023]
【Example】
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0024]
(Example 1)
FIG. 2 is a connection diagram showing the configuration of the network speech recognition system according to the first embodiment of the present invention. A group of electric devices installed in a certain home is provided with a microphone 1 as a means for converting sound into an electric signal, a speaker 2 for restoring a synthesized sound signal into sound, and a connecting means 3 for connecting to a network. ing.
[0025]
Therefore, for example, the operation unit of the laundry 4 only has the microphone 1, the speaker 2, and the display window other than the microphone 1. There is no timer or keyboard as in the related art. Note that the connection means 3 is built in the device body.
[0026]
The microwave oven 5, the rice cooker 6, and the refrigerator 7 are exactly the same, and the operation unit has only the microphone 1, the speaker 2, and other display windows as necessary. The television 8 is similar, but the television originally has the speaker 2 as an original function.
[0027]
Now, there is a central processing unit 9 connected to these devices via a line via the connection unit 3. This controls not only a home network but also a large number of home electrical appliances via a line. The voice recognition means 10 is built in.
[0028]
FIG. 1 is a block diagram showing the configuration of the network speech recognition system according to the first embodiment of the present invention. The laundry 4 is provided with a control unit 11 in addition to the microphone 1, the speaker 2, and the connection unit 3 described above, and controls the motor and the heater 12 based on a voice command. That is, the control unit 11 transmits the operation command input from the microphone 1 to the central processing unit 9 via the connection unit 3 and the line. The voice data passes through the connection means 14 provided in the central processing means 9, and is identified by the ID recognizing means 15 at which location (home) and which address (electric equipment), and recognizes that the laundry is a home laundry A. After that, the instruction content is analyzed by the voice recognition means 10. In the speech recognition, the similarity with the speech pattern recorded in the storage unit 16 is compared, and the sentence is estimated from the preceding or succeeding words or phonemes. As a result of the recognition process, what operation the device should take is transmitted to the laundry 4 again. In response to this output, the control means 11 starts, changes, or stops energization of the motor and the heater 12. The control means 11 outputs the synthesized voice message transmitted from the speaker 2 as needed.
[0029]
The same applies to microwave ovens 5 and below. In the microwave oven 5, what is controlled by the control means 11 is a magnetron 13 which is a heat source. Of course, whether the laundry 4 or the microwave oven 5 is controlled, the object to be controlled is not limited to this. These are the controlled main blocks. The output from the central processing means 9 may be data for directly controlling a predetermined controlled block or data obtained by coding a certain operation mode. These are once decoded by the control means 11 and control the controlled block.
[0030]
With such a configuration, the network voice recognition system makes it possible for the operator and the device to interact with each other via a network, such as a heating device having a voice recognition unit and a voice synthesis unit as disclosed in Japanese Patent Application Laid-Open No. 56-88503. The operation can be advanced as if it were, and the operation of the device can be executed reliably. Of course, speech synthesis is not essential to the present invention.
[0031]
(Example 2)
FIG. 3 is a connection diagram showing the configuration of the network speech recognition system according to the second embodiment of the present invention. Each home, such as a certain home A17 and a certain home B18, is connected to the central processing means 9 via the Internet. Each of the electric appliances installed in a certain home A17 is provided with a microphone as a means for converting sound into an electric signal and a speaker for restoring a synthesized sound signal into sound as in FIG. However, the connection means for connecting to the network does not have individual electric devices. These devices are connected to a hub switch 19, which is connected to a router. That is, a single router shares the connection means with the central processing means 9. The inside of the home B18 has the same configuration.
[0032]
With such a configuration, the present invention can be economically utilized by sharing an expensive modem at present. However, in preparation for the realization of Internet home appliances in the future, the development of an inexpensive OS that simplifies the communication protocol and connects to the Internet that can be mounted on an 8-bit microcomputer has been advanced, and is more expensive than the connected electrical equipment. Some modems will be a funny story in a few years. In short, the configuration shown in FIG. 3 is realistic for the time being, but the configuration shown in FIG. 1 will be general in the future.
[0033]
This selection is a comparison between the cost of connecting to the network (hardware and software such as usage fee) and the hardware cost of the voice recognition means. If the former is greater than the latter, the invention will actually have no users. However, as will be described later, the speech recognition processing has high-speed arithmetic processing and a large amount of storage means, and there is a technological innovation of rapid progress.According to the present invention, these can be shared by the central processing means and while updating to the latest technology, Very simple control means for controlling the operation of the heat source, the motor and the like may be provided in each electric device. In some cases, all the control can be performed by the central processing means. It is a premise of the present invention that network costs are steadily falling and voice processing technology is evolving.
[0034]
(Example 3)
FIG. 4 is a block diagram showing the configuration of the network voice recognition system according to the third embodiment of the present invention. In contrast to the first embodiment shown in FIG. 1, an A / D converter 21 is connected to the microphone 1, the audio is converted to a digital signal, and the control unit 11 is connected via the connection unit 3 to the central processing unit 9 via a line. Send to The audio signal transmitted by such a configuration is resistant to noise. That is, if the transmission data is divided into several frames and an even or odd parity check is performed for each frame, noise can be easily eliminated. If a digital signal is used, processing such as compression can be easily performed. The A / D converter is inexpensive, and a configuration in which the A / D converter is individually mounted on each electric device is useful in practical use in terms of increasing reliability.
[0035]
In this embodiment, when the central processing unit 9 needs to transmit a message to the electric device based on the result of the voice recognition processing, a predetermined voice is synthesized from the storage unit 16 and transmitted to the electric device via the line. This is a configuration for transmission. Further, each electric device includes a D / A converter 22. The voice synthesized by the central processing unit 9 may be transmitted over the line as analog data. However, if it is handled as digital data, it is strong against noise and can easily perform processing such as compression. That is, it is possible to obtain the same effect as digitizing a voice command in voice recognition.
[0036]
(Example 4)
FIG. 5 is a block diagram showing a configuration of the voice recognition processing of the network voice recognition system according to the fourth embodiment of the present invention.
[0037]
Speech uttered by humans is extremely vague, especially in Japanese, where the independence between words is poor, and there are many factors that hinder accurate recognition, such as individual differences and regional characteristics (dialects). Meanwhile, the voice recognition processing is realized by making full use of pattern recognition technology. There are specific speaker recognition in which words to be recognized are registered in advance, and unspecified speaker recognition in which the voices of an unspecified number of people are recognized. Naturally, the latter is technically high, but practical. .
[0038]
The speech recognition includes word unit recognition and phoneme unit recognition. In word-based recognition, a computer analyzes speech, extracts features, and creates a time series of feature quantities. Then, it compares and calculates the similarity with the characteristic time-series word dictionary in the processing means and outputs it as a recognition result.
[0039]
Recognition of phoneme units converts input speech into a phoneme symbol string and replaces it with a word string. Parse this and convert it to a string. Furthermore, it performs sentence analysis and semantic analysis to generate sentences. By recognizing the situation before and after the sound is uttered and the individual devices connected to the central processing means, it is possible to go to a simple language understanding such as a possible command and a impossible command. It is extremely difficult, so a high-speed, large-capacity computer / dedicated hardware hybrid system works efficiently.
[0040]
In this embodiment, a method based on HMM (Hidden Markov Model) that is strong against fluctuation of an unspecified speaker is adopted. HMM is a technique widely used in recent years, and realizes speech recognition in phoneme units based on this technique. Such a technology is described in, for example, "The Unspecified Speaker Phoneme Level Speech Recognition / Learning Circuit Based on Hidden Markov Model (Kazuhiro Nakamura et al., Nara Institute of Science and Technology)," which won the IP Award of the 2nd IP Award in 2000. Is shown.
[0041]
In FIG. 5, the audio electric signal (analog) is replaced by the A / D converter 21 with a digital signal. The A / D converter 21 may be provided in each electric device or in the central processing means. The effect when placed in the apparatus has already been described in FIG.
[0042]
The audio data converted into the digital signal is sent to the controller 23 and is sequentially stored in the LPC memory 24. At the same time, the data is sent to the LPC analysis unit 25 one frame at a time, and speech features are extracted using a linear prediction analysis (LPC) technique. That is, a feature vector is calculated. The feature vector is obtained by calculating a linear prediction coefficient based on the time series data of the voice and the autocorrelation function. The obtained feature vector is recorded in the LPC memory.
[0043]
Next, in the VQ unit 26, the feature vector sequence obtained by the LPC analysis is converted into a finite number of symbols. Vector quantization (VQ). For each feature vector, a codebook (a table having a pair of a symbol number and a representative vector as an element) is searched and mapped to a symbol of the representative vector having the shortest distance. The codebook is generated by applying a clustering method to a large amount of learning data, and is stored in the LPC memory 24.
[0044]
Subsequent to the above processing, speech recognition is performed by the HMM unit 27. The HMM is a phoneme probability model (probability automaton). One HMM is configured for each phoneme, and a probability parameter is stored in the HMM memory 28. The phoneme showing the symbol sequence and the highest probability is selected. Then, the result of this recognition processing is output.
[0045]
In this embodiment, the voice recognition is performed by hardware, but most of the processing is a matrix calculation and a probability calculation process that are repeated endlessly. If the central processing means has a sufficient processing capability, the LPC analysis unit, VQ unit, HMM unit, etc. can be replaced by software processing by a program. In addition, a hybrid system of hardware and software can be used to balance processing speed and economy.
[0046]
(Example 5)
FIG. 6 is a flowchart of a program having a configuration in which the central processing unit according to the fifth embodiment of the present invention performs program processing.
[0047]
First, digitized audio data is input (step 100). The device ID data is added before or after the voice data. The device recognizes the ID data and specifies the device to which the voice data has been sent (step 101). The ID is given as a unique ID when the electrical device is shipped from the factory. When the electrical device is installed in a certain home, the individual information of the home is automatically or automatically registered. And stored in the central processing means. By accessing such a database, the central processing means can specify the accessed home and the configuration of the network device in the home (step 102).
[0048]
Subsequently, the voice recognition process starts. First, voice data is divided for each frame and features are extracted (step 103). Then, the feature vector is quantized (step 104), and the probability that the data is close to each other is calculated for each phoneme (step 105). Then, the HMM with the highest probability is obtained (step 106). The input voice is determined, and the voice recognition process ends (step 107). The voice recognition process during this period is exactly the same as that executed in hardware in the example of FIG.
[0049]
The determined voice is searched by searching a voice list (text table) registered for each device, and matching is performed (step 108). This is to determine whether or not the command is a command accepted by the device. For example, since the ID identifies the type of microwave oven of which manufacturer and from which the audio data was input, the inspection is performed in light of the functions and specifications of the microwave oven. A microwave oven can accept a command of "baking at 180 ° C. for 35 minutes in an oven", but a microwave oven without a heater function cannot execute this command. Also, even if a command to "select automatically" is received in the microwave oven, it cannot be handled. This voice list (text table) may be stored in the memory of the central processing unit every time a new product is released, or the central processing unit may search for data via a network to a server owned by the manufacturer. You may. First, it is checked whether the voice is a registered voice in the table (step 109). If the voice is not found in this search, an error message is output as voice (step 110). A message can be issued in detail according to the nature of the error.
[0050]
If the voice is a registered voice, the stage (operating state) of the device is checked (step 111). All commands are instructed by voice to the central processing means so that the central processing means can specify what state the device is in now. Then, even if there is no mistake in the voice data, it is checked whether the voice data can be accepted now (step 112). If the user is instructed to "stop" while the microwave oven is not operating, he will not be able to respond. Also, it cannot respond during execution of the instruction received immediately before.
[0051]
Further, since the central processing unit knows the configuration of the home network device, the operation of other devices connected to the network can be checked (step 113). For example, when the rice cooker has just started operation, and it has been cooked 40 minutes later, and the microwave oven is instructed to reheat it, ask the operator by sounding out whether it is acceptable to accept it and calling for the instruction again. Is also possible.
[0052]
After such processing, a sound is output to notify that the reception has been completed (step 114), and an operation output for causing the device to perform a predetermined operation is retrieved from the table (step 115). For example, a microwave oven has loads such as a magnetron, a motor, and a lamp as a heat source. An operation state chart of which of these loads is turned on and which is turned off is stored in advance as a table. When these states are handled by four-digit hexadecimal data, it is necessary to determine in advance the equipment side and the central processing means that the 01A6 operates the magnetron at full output, the turntable motor rotates, and the lamp is turned off. Is output as an operation command (step 116).
[0053]
With the above program, the central sled means can control the electric equipment connected to the net by voice recognition.
[0054]
【The invention's effect】
As described above, according to the present invention, a plurality of electric devices connected to a network can be operated by a human-friendly operation system called voice recognition, and a high recognition rate is always maintained by using the latest voice recognition technology. However, a system that can be used economically can be realized.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a network speech recognition system according to a first embodiment of the present invention.
FIG. 2 is a connection diagram showing a configuration of the network speech recognition system.
FIG. 3 is a connection diagram showing a configuration of a network speech recognition system according to a second embodiment of the present invention.
FIG. 4 is a block diagram showing a configuration of a network speech recognition system according to a third embodiment of the present invention.
FIG. 5 is a block diagram showing a configuration of a voice recognition process of a network voice recognition system according to a fourth embodiment of the present invention.
FIG. 6 is a flowchart of a program having a configuration in which a central processing unit according to a fifth embodiment of the present invention performs program processing;

Claims (9)

An electrical device having means for converting voice into an electrical signal, and a central processing unit connected to the device by a line, the central processing unit performing voice recognition processing of the transmitted voice signal, and processing results A network voice recognition system configured to transmit a predetermined operation output to the electric device based on the information. The network voice recognition system according to claim 1, comprising a plurality of the electric devices. The network voice recognition system according to claim 1, wherein the electric device has an individual ID. 2. The electric apparatus according to claim 1, wherein the electric device including the means for converting the sound into an electric signal further includes an A / D converter, and converts the sound into a digital signal and transmits the digital signal to the central processing means via a line. Network speech recognition system. The network speech recognition system according to claim 1, wherein the speech recognition processing performed by the central processing means is a speech recognition processing based on linear prediction analysis (LPC), vector quantization (VQ), and a Hidden Markov Model (HMM). 2. The network according to claim 1, wherein the central processing unit synthesizes a predetermined voice from a memory and transmits the synthesized voice via a line when a message needs to be transmitted to the electric device based on a voice recognition processing result. Voice recognition system. 2. The network speech recognition system according to claim 1, wherein said central processing means has hardware for performing speech recognition and a program for performing software processing. An electrical device converting voice to an electrical signal; an electrical device transmitting an audio signal; a central processing unit performing a voice recognition process on the transmitted audio signal; and Transmitting a predetermined operation output to the network voice recognition method. An electric device converting sound into an electric signal; a step in which the electric device transmits a sound signal; a step in which a central processing unit performs a sound recognition process on the transmitted sound signal; and And transmitting a predetermined operation output to the network voice recognition program.

Images