JP5263875B2 - Computer system, speech recognition method and computer program for speech recognition - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a computer system for voice recognition of utterance input, capable of improving recognition accuracy of voice recognition. <P>SOLUTION: The system includes: a first determination section for determining whether or not, in response to input of first utterance, voice of the input first utterance matches voice registered in a storage section; a request section for requesting input of second utterance, when the voice of the input first utterance does not match the voice registered in the storage device; a second determination section for determining whether or not, the voice of the input second utterance matches the voice registered in the storage device; a comparing section for comparing a phoneme string of the second utterance with a phoneme string of the first utterance, when the voice of the second utterance matches the voice registered in the storage section; and a relating section for relating the voice of the first utterance to a command or action corresponding to the second utterance, when the phoneme string of the second utterance is similar to the phoneme string of the first utterance. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a computer system for speech recognition of speech input, a method thereof, and a computer program.

As an in-vehicle device (for example, a car navigation system) is highly functional and complicated, speech recognition is known as a human interface that is easy for a user to use. When the user issues a voice command to the in-vehicle device, the voice command is converted into a command through voice recognition, and an action corresponding to the command is executed. Here, when the user uses a function (hereinafter referred to as a voice tag registration mode) for associating an arbitrary utterance content with an in-vehicle device command, new utterance registration is performed according to the following procedure:
1. Activate voice tag registration mode,
2. Menu selection to associate utterances with commands,
3. Utterance,
4). sign up.
However, the user is not always aware of which command is associated with which utterance. In addition, when the voice recognition fails, the user does not notice the failure of the voice recognition for the first time or does not notice the failure. Furthermore, in the environment of the in-vehicle device, the user does not want to perform the work of registering the utterance content or correcting the pronunciation. Furthermore, it is troublesome for the user to execute the above procedure each time in order to associate the utterance content with the command.

  Patent Document 1 describes a recognition dictionary system that is a terminal used by a user and includes a client terminal that includes a recognition dictionary that stores character string information, and a server terminal that is connected to the client terminal via a communication network. The server terminal cumulatively stores a paraphrase vocabulary having a meaning equivalent to the second vocabulary stored in the recognition dictionary and having different character string information for the first vocabulary corresponding to the information input from the client terminal. A paraphrasing vocabulary accumulating unit and a paraphrasing vocabulary accumulating unit, analyzing the occurrence frequency of the paraphrase vocabulary and determining at least one paraphrase vocabulary having the occurrence frequency higher than a first predetermined value as a main paraphrase vocabulary A frequency analysis unit, and a recognition dictionary update unit that updates the recognition dictionary in order to register the first vocabulary determined as the main paraphrase in the recognition dictionary in association with the second vocabulary.

JP 2007-213055 A

There is a need to increase the recognition accuracy of voice recognition in in-vehicle devices. Therefore, in order to increase the recognition accuracy, for example, a voice recognition system that adds a variation of pronunciation of a word or a phrase and automatically responds to the fluctuation of the sound is required.
Also, the work of adding variations in the pronunciation of words or phrases tends to be avoided in terms of cost because the accuracy cannot be improved without manual intervention. Accordingly, there is a need for a speech recognition system that automatically registers variations of pronunciation of words or phrases.
Furthermore, hardware resources may be limited in in-vehicle devices. Nevertheless, for example, when associating multiple sound fluctuations with a single word or phrase, more memory capacity is required. Therefore, there is a need for a speech recognition system that can cope with hardware resource constraints.

The present invention provides a computer system for speech recognition of speech input.
The computer system is
In response to the input of the first utterance, a first determination unit that determines whether the input voice of the first utterance matches the voice registered in the storage unit;
A request unit for requesting input of a second utterance when the input voice of the first utterance does not match the voice registered in the storage unit;
A second determination unit that determines whether or not the voice of the input second utterance matches the voice registered in the storage unit;
A comparison unit that compares the phoneme sequence of the second utterance and the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit;
An association unit for associating the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance;

  In one embodiment of the present invention, the computer system further includes a first generation unit that generates a first phoneme string corresponding to the first utterance together with the first determination.

  In one embodiment of the present invention, the first determination unit further determines whether or not the generated first phoneme string matches a phoneme string registered in the storage unit.

  In one embodiment of the present invention, whether or not the input speech of the first utterance matches the speech registered in the storage unit is determined based on the degree of coincidence of phonemes in the phoneme string.

  In one embodiment of the present invention, the computer system executes an action corresponding to the first utterance when the inputted voice of the first utterance matches the voice registered in the storage unit. An execution part is further included.

  In one embodiment of the invention, the first utterance is a first voice command. In one embodiment of the present invention, the execution unit executes an action corresponding to the first voice command when the first voice command matches the voice command registered in the storage unit.

  In one embodiment of the present invention, the computer system stores the voice of the first utterance in the recording unit when the inputted voice of the first utterance does not match the voice registered in the storage unit. A recording unit is further included.

  In one embodiment of the present invention, the computer system further includes a second generation unit that generates a second phoneme string corresponding to the second utterance together with the second determination.

  In one embodiment of the present invention, in the computer system, the second determination unit further determines whether or not the generated second phoneme sequence matches a phoneme sequence registered in the storage unit. To do.

  In one embodiment of the present invention, whether or not the voice of the input second utterance matches the voice registered in the storage unit is determined based on the matching degree of phonemes in the phoneme string.

  In one embodiment of the present invention, the computer system executes an action corresponding to the second utterance when the inputted voice of the second utterance matches the voice registered in the storage unit. A second execution unit is further included.

  In one embodiment of the invention, the second utterance is a second voice command. In one embodiment of the present invention, the second execution unit performs an action corresponding to the second voice command when the second voice command matches the voice command registered in the storage unit. Run.

  In one embodiment of the present invention, it is determined based on the degree of coincidence of phonemes in the phoneme sequence that the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance.

  In one embodiment of the present invention, when the phoneme string of the second utterance is similar to the phoneme string of the first utterance, it is determined based on the sound fluctuation information whether or not the voices match.

  In one embodiment of the present invention, the computer system registers the phoneme sequence of the first utterance in the storage unit when the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance. It further includes a registration unit that determines whether or not to do so.

  In one embodiment of the present invention, the registration unit refers to a policy for determining whether to register the phoneme string of the first utterance in the storage unit.

  In one embodiment of the present invention, the policy is based on at least one of a high noise ratio, a frequency of use of words or phrases, and a sequence of phoneme strings.

  In one embodiment of the present invention, the computer system further includes a deletion unit that determines whether or not to delete the phoneme string of the registered utterance from the storage unit.

  In one embodiment of the present invention, the computer system corresponds to the second utterance when the phoneme sequence of the second utterance is not similar to the phoneme sequence of the first utterance. A query unit for querying the user as to whether or not to associate with the command or action to be performed.

  In one embodiment of the present invention, the association unit associates the voice of the first utterance with the command or action corresponding to the second utterance in response to receiving an instruction to perform the association by the user.

  In one embodiment of the present invention, the computer system performs the second utterance on the voice of the first utterance when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance. And a selection unit that allows selection of a command or action corresponding to.

  In one embodiment of the invention, allowing the selection includes presenting a list of voice commands.

  In one embodiment of the present invention, associating the voice of the first utterance with a command or action corresponding to the second utterance registers the voice of the first utterance as a variation of the voice of the second utterance. including.

In one embodiment of the invention, the computer system comprises:
A second request unit that further requests input of a third utterance when the input second utterance speech does not match the voice registered in the storage unit;
A third determination unit for determining whether or not the voice of the input third utterance matches the voice registered in the storage unit;
A second comparison unit that compares the phoneme sequence of the third utterance and the phoneme sequence of the second utterance when the speech of the third utterance matches the speech registered in the storage unit;
A second associating unit for associating the voice of the second utterance with the command or action corresponding to the third utterance when the phoneme string of the third utterance is similar to the phoneme string of the second utterance; .

The present invention also provides a method for speech recognition of speech input. The method causes a computer system to perform the following steps.
The step is
In response to an input of the first utterance, a first determination step of determining whether or not the input voice of the first utterance matches the voice registered in the storage unit;
Requesting the input of the second utterance when the input voice of the first utterance does not match the voice registered in the storage unit;
A second determination step of determining whether or not the voice of the input second utterance matches the voice registered in the storage unit;
Comparing the phoneme sequence of the second utterance with the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit;
Associating the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance.

  In one embodiment of the present invention, the first determination step further includes a step of generating a first phoneme string corresponding to the first utterance together with the first determination.

  In one embodiment of the present invention, the first determining step further includes a step of further determining whether or not the generated first phoneme string matches a phoneme string registered in the storage unit.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of executing an action corresponding to the first utterance when the inputted voice of the first utterance coincides with the voice registered in the storage unit.

  In one embodiment of the invention, the first utterance is a first voice command. In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of executing an action corresponding to the first voice command when the first voice command matches the voice command registered in the storage unit.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of storing the voice of the first utterance in the recording unit when the inputted voice of the first utterance does not match the voice registered in the storage unit.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of generating a second phoneme string corresponding to the second utterance together with the second determination.

  In one embodiment of the present invention, the second determining step includes a step of further determining whether or not the generated second phoneme string matches the phoneme string registered in the storage unit.

  In one embodiment of the invention, the second utterance is a second voice command. In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of executing an action corresponding to the second utterance when the inputted voice of the second utterance matches the voice registered in the storage unit.

  In one embodiment of the invention, the second utterance is a second voice command. In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of executing an action corresponding to the second voice command when the second voice command matches the voice command registered in the storage unit.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of determining whether or not to register the voice of the first utterance in the storage unit when the phoneme string of the second utterance is similar to the phoneme string of the first utterance.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes a step of determining whether or not to delete the phoneme string of the registered utterance from the storage unit.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step determines whether to associate the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance does not resemble the phoneme string of the first utterance. Including the step of inquiring.

  In one embodiment of the present invention, the step of associating associates the voice of the first utterance with the command or action corresponding to the second utterance in response to receiving an instruction to perform the association by the user. Includes steps.

  In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step includes selecting a command or action corresponding to the second utterance for the voice of the first utterance when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance. Includes forgiving steps.

  In one embodiment of the invention, allowing the selection includes presenting a list of voice commands.

  In one embodiment of the present invention, the step of associating the voice of the first utterance with the command or action corresponding to the second utterance registers the voice of the first utterance as a variation of the voice of the second utterance. including.

In one embodiment of the invention, the method further causes the computer system to perform the following steps: The step is
Further requesting the input of the third utterance when the inputted voice of the second utterance does not match the voice registered in the storage unit;
Determining whether the voice of the input third utterance matches the voice registered in the storage unit;
Comparing the phoneme sequence of the third utterance and the phoneme sequence of the second utterance when the speech of the third utterance matches the speech registered in the storage unit;
Associating the voice of the second utterance with the command or action corresponding to the third utterance when the phoneme string of the third utterance is similar to the phoneme string of the second utterance.

The present invention also provides a method for speech recognition of speech input. The method causes a computer system to perform the following steps.
The step is
In response to the input of the first utterance, determining whether the input voice of the first utterance matches the voice registered in the storage unit;
Executing the action corresponding to the first utterance when the inputted voice of the first utterance matches the voice registered in the storage unit;
Requesting the input of the second utterance when the input voice of the first utterance does not match the voice registered in the storage unit;
Determining whether the voice of the input second utterance matches the voice registered in the storage unit;
Comparing the phoneme sequence of the second utterance with the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit;
Associating the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance;
If the phoneme sequence of the second utterance is not similar to the phoneme sequence of the first utterance, inquiring the user whether to associate the voice of the first utterance with the command corresponding to the second utterance;
When the voice of the second utterance does not match the voice registered in the storage unit,
Further requesting input of a third utterance;
Determining whether the voice of the input third utterance matches the voice registered in the storage unit;
Comparing the phoneme sequence of the third utterance and the phoneme sequence of the second utterance when the speech of the third utterance matches the speech registered in the storage unit;
Associating the voice of the second utterance with the command or action corresponding to the third utterance when the phoneme string of the third utterance is similar to the phoneme string of the second utterance.

  The present invention also provides a computer program for speech recognition of speech input. The computer program causes a computer system to execute the steps described in any one of the above methods.

  The computer system according to the embodiment of the present invention additionally registers different phoneme strings at the time of failure as the same command with respect to the voice command at the time of successful speech recognition by simultaneously performing speech recognition and phoneme acquisition a plurality of times. As a result, the process of additionally registering different phoneme sequences at the time of failure can be performed simultaneously with the voice recognition, so that the user's voice registration work is simplified. Further, the user does not need to worry about what kind of utterance is registered for a certain voice command. In addition, since the computer system according to the embodiment of the present invention can additionally register a variation word or phrase according to the user after the computer system is shipped, the speech recognition accuracy is improved. Since words or phrases can be additionally registered, it is possible to reduce the capacity of the dictionary for speech recognition when the computer system according to the embodiment of the present invention is shipped.

In the embodiment of the present invention, “utterance” refers to an utterance input by a user to a computer system. “Speech input” is input through, for example, a voice input unit connected to a computer system, such as a microphone or a sound card.
In the embodiment of the present invention, the “storage unit” includes data on speech and / or phoneme strings. The “storage unit” may be a database. The storage unit may be disposed in a storage device in the computer system or outside the computer system, or a storage device of a server or proxy connected to the computer system via a network.
In the embodiment of the present invention, a conventional technique can be used as the method of “voice recognition”. For example, IBM Embedded ViaVoice (EVV) can be used as a speech recognition technology. In speech recognition, a recognition process, a word or phrase dictionary, and an acoustic model are used to perform word or phrase recognition processing on an input utterance.
In the embodiment of the present invention, the “phoneme string” refers to a minimum sound unit used for phonological theory to distinguish (discriminate) meaning in an arbitrary individual language. Phonemes are described by enclosing them with //. The symbols used for phonemes are arbitrary, and each language's phoneme characters may be used, or the international phonetic alphabet may be used.
In the embodiment of the present invention, a conventional technique can be used as a method for comparing phoneme strings. For example, an API that compares phoneme sequences in IBM EVV can be used to determine whether phoneme sequences are similar, for example. An API for comparing phoneme strings is, for example, esrCompareBaseforms.
In the embodiment of the present invention, “sound matches” means that a word or phrase in a recognition result is a word in a dictionary (a set of words or phrases in a storage area) with a value (score) that is equal to or higher than a predetermined threshold value (score). Or to match a phrase. The predetermined threshold may vary depending on the language. Also, it can be determined that the voices coincide with each other by executing an action corresponding to the voice.
In the embodiment of the present invention, “sound does not match” means that a word or phrase not less than or exceeding the predefined threshold value is not found in the dictionary. Further, it is possible to determine that the voices do not coincide with each other because the action corresponding to the voice has not been executed.
In the embodiment of the present invention, that phoneme strings are similar means that the degree of coincidence of phonemes in the phoneme string is equal to or higher than a predefined threshold. The predetermined threshold may vary depending on the language.
In the embodiment of the present invention, that phoneme strings are not similar means that the degree of coincidence of phonemes in the phoneme string is equal to or lower than a predetermined threshold value.
In the embodiments of the present invention, the “computer system” includes an in-vehicle device such as a car navigation system, a handheld computer, a personal digital assistant, a mobile phone or a navigation system other than a car navigation system. Not limited to.

  Embodiments of the present invention will be described below with reference to the drawings. It should be understood that this embodiment is for the purpose of illustrating a preferred aspect of the present invention and is not intended to limit the scope of the invention to what is shown here. Further, throughout the following drawings, the same reference numerals refer to the same objects unless otherwise specified.

FIG. 1A shows an outline of a speech recognition system which is an embodiment of the present invention.
The speech recognition system (101) can be a navigation system, a handheld computer, a personal digital assistant, or a mobile phone.
The speech recognition system (101) includes a recognition processing unit (102), a determination unit (103), a storage unit (104), a request unit (105), a comparison unit (106), an association unit (107), and an inquiry unit (108). ), A selection unit (109), a registration unit (110), a policy (111), an execution unit (112), and a deletion unit (113).
The recognition processing unit (102) includes, for example, a recognition engine, a word or phrase dictionary, and an acoustic model. When a speech is input through a speech input unit (not shown), for example, a microphone or a sound card microphone, the recognition engine performs speech recognition processing of a word or phrase on the input speech, a word or phrase dictionary, and an acoustic model. To generate a phoneme sequence at the same time. The recognition process is to perform matching between an utterance and a word or phrase. The recognized speech and the generated phoneme string are associated, for example, by generating a phoneme string.
The determination unit (103) determines whether the audio data processed in the recognition processing unit (102) matches the audio data stored in the storage unit (104). Specifically, when the user inputs a first utterance, for example, a voice command, the determination unit (103) determines whether the phoneme sequence of the voice command matches the phoneme sequence registered in the storage unit (104). to decide. Similarly, when the user inputs a second or subsequent utterance, for example, a voice command, the determination unit (103) determines whether the phoneme string of the voice command matches the phoneme string registered in the storage unit (104). Judging.
The storage unit (104) stores voice data and phoneme string data for use in the determination unit (103). The storage unit (104) can be, for example, a database.
The request unit (105) inputs the second utterance (second utterance) when the voice of the first utterance (first utterance) input by the user does not match the voice registered in the storage unit (104). Request. Similarly, when the voice of the utterance (second utterance) input by the user for the second time does not match the voice registered in the storage section (104), the request unit (105) Utterance) input. Thereafter, the request unit (105) requests the user to input an utterance for a preset number of times.
The comparison unit (106) compares the phoneme sequence of the second utterance with the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit (104). Similarly, the comparison unit (106) compares the phoneme sequence of the nth utterance and the phoneme sequence of the (n-1) th utterance when the speech of the nth utterance matches the speech registered in the storage unit (104). To do.
The associating unit (107) associates the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance. Similarly, for example, when the phoneme sequence of the n-th utterance is similar to the phoneme sequence of the n-th or earlier utterance, the associating unit (107) Associate with command corresponding to n utterances.
When the phoneme sequence of the second utterance is not similar to the phoneme sequence of the first utterance, the inquiry unit (108) inquires of the user whether to associate the voice of the first utterance with the command corresponding to the second utterance. Similarly, when the phoneme sequence of the nth utterance does not resemble the phoneme sequence of the n-1st utterance, the inquiry unit (108) associates the voice of the n-1st utterance with the command corresponding to the nth utterance. Ask the user if it is.
The selection unit (109) allows a command corresponding to the voice of the first utterance to be selected when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance. Similarly, for example, when the phoneme sequence of the n-th utterance is similar to the phoneme sequence of the n-th or earlier utterance, the selection unit (109) is a command corresponding to the n-th or earlier speech. Allows you to choose.
When the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance, the registration unit (110) registers the speech data of the first utterance, particularly the phoneme sequence of the first utterance, in the storage unit (104). Whether or not the voice data of the first utterance is registered in the storage unit (104) is determined according to the policy.
The policy (111) is based on, for example, high noise ratio, frequency of use of words or phrases, or arrangement of phoneme strings, but is not limited thereto.
The execution unit (112) executes an action corresponding to the first utterance when the input voice of the first utterance matches the voice registered in the storage unit (104). In particular, the execution unit (112) executes an action corresponding to the first voice command when the input first voice command matches the voice registered in the storage unit (104). Similarly, the execution unit (112) executes an action corresponding to the nth utterance when the input nth utterance speech matches the speech registered in the storage unit (104). In particular, the execution unit (112) executes an action corresponding to the nth voice command when the inputted nth voice command matches the voice registered in the storage unit (104).
The action differs depending on the product on which the speech recognition system (101) is installed. For example, when the voice recognition system (101) is a car navigation system, the action displays, for example, a window for search, displays a route to a predetermined destination according to a search keyword, DVD playback function or music playback function , Turn on / off the power of the car navigation system, but is not limited thereto.
The deletion unit (113) allows the user to delete the voice data or phoneme string data registered in the storage unit (104). The reason for deleting is to limit the capacity of the storage unit for storing the voice data or the phoneme string data, and to prevent the recognition rate from decreasing due to the excessive increase of phoneme string variations.
The target of deletion is voice data or phoneme string data that is used less frequently. The voice recognition system (101) may include a presentation unit that presents voice data or phoneme string data with low usage frequency as a list to the user. The presentation may simultaneously display on the display the date when the voice command was last used and the number of times the voice command was used, together with the voice command that may be deleted. The display is displayed in a window format, for example. Alternatively, instead of displaying on the display a voice command that may be a deletion target, it may be performed while confirming with the user whether or not the voice command is reproduced and deleted.

FIG. 1B shows a conceptual diagram of actions for voice, phonemes, and voice commands, which is an embodiment of the present invention.
A voice command is input to the voice input unit (122) of the voice recognition system (FIG. 1A, 101). The recognition processing unit (123) performs speech recognition as a word or phrase and phoneme string recognition on the input speech. In the storage unit (124), for example, in the memory, the voice command “use highway” recognized by the recognition processing unit (123) is stored. The execution unit (FIG. 1A, 112) confirms whether a command corresponding to the recognized voice command exists in the command group (125). The command group is stored in advance in a predetermined storage unit, for example, a database. When a command corresponding to the recognized voice command is found in the command group, an action “display a route using the highway to the destination” corresponding to the voice command “use highway” is displayed. 126) and execute the action.

FIG. 2A shows an outline (part 1) of a method for speech recognition, which is an embodiment of the present invention.
In step 201, the user inputs the first utterance (first utterance) through the voice input unit. The speech recognition system (FIG. 1A, 101) starts speech recognition. The start of speech recognition may be before or after the first utterance is input. The input utterance is converted into an electrical voice signal and passed to the recognition engine. The recognition engine recognizes a word or phrase with respect to the input speech signal and simultaneously generates a phoneme string. The recognized word or phrase for the first utterance and the generated phoneme sequence are stored in memory or a hard disk drive or solid state disk.
For example, when a user uses an expressway and wants to go to a destination, the user utters “Corso Doroyo” to the car navigation system.
In step 202, in response to the input of the first utterance, the voice recognition system determines whether or not the input voice of the first utterance matches the voice registered in the storage unit.
In the above example, it is assumed that the first utterance is “use highway” and the phoneme sequence of the first utterance is “koo: sokudoo: roo: riyoo:”. On the other hand, it is assumed that the phoneme string registered in the storage unit is “kousokudourooriyou”. In this example, since the phoneme sequence of the first utterance is recognized as a long note “-” for “u” or is not recognized, the phoneme sequence “koo: sokudoo: roo: riyoo” of the first utterance is recognized. : "And the phoneme string" kousokudourooriyou "registered in the storage unit are determined not to match.
In step 203, the speech recognition system (101) requests the user to input the second utterance (second utterance) if there is no match. The request can be displayed, for example, on a display device connected to a voice guidance or voice recognition system. The voice or phoneme string of the first utterance is stored in a memory, a hard disk drive, or a solid state disk.
For example, the user utters “Korokudourooriyo” to the car navigation system.
In step 204, the speech recognition system (101) executes an action corresponding to the first utterance when there is a match.
In step 205, the user inputs the second utterance through the voice input unit. The input utterance is converted into an electrical voice signal and passed to the recognition engine. The recognition engine converts the input utterance into an electrical voice signal and passes it to the recognition engine. The recognition engine recognizes a word or phrase with respect to the speech signal and simultaneously generates a phoneme string. The recognized word or phrase for the second utterance and the generated phoneme sequence are stored in memory or a hard disk drive or solid state disk.
In the above example, it is assumed that the second utterance is “use highway” and the phoneme string of the second utterance is “kousokudourooriyou”.
In step 206, in response to the input of the second utterance, the voice recognition system (101) determines whether or not the input voice of the second utterance matches the voice registered in the storage unit.
In the above example, the phoneme string “kousokudourooriyou” of the second utterance matches the phoneme string “kousokudourooriyou” registered in the storage unit.
In step 207, when there is a match, the speech recognition system (101) compares the phoneme sequence of the second utterance with the phoneme sequence of the first utterance.
In the above example, the phoneme sequence “kousokudourooriyou” of the second utterance is compared with the phoneme sequence “koo: sokudoo: roo: riyoo:” of the first utterance.
In step 208, the speech recognition system (101) further executes an action corresponding to the second utterance if there is a match.
In this example, the phoneme sequence “kousokudourooriyou” of the second utterance matches the phoneme sequence “kousokudourooriyou” registered in the storage unit. Therefore, it is determined that the phoneme string of the second utterance matches the phoneme string registered in the storage unit. Therefore, as an action corresponding to “use highway”, for example, a route using the highway to the destination is displayed on the display device of the car navigation system.
In step 209, if the voice recognition system (101) does not have the above match, the process proceeds to the step shown in FIG. 2B.
In step 210, the speech recognition system (101) associates the speech of the first utterance with the command or action corresponding to the second utterance if the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance. In particular, the first phoneme string is associated with a command or action corresponding to the second voice command. This association makes it possible to execute an action corresponding to the second voice command by performing the same utterance as the first voice command thereafter. Therefore, the same utterance as that of the second voice command is not required thereafter.
In the above example, the phoneme sequence “kousokudourooriyou” for the second utterance is compared to the phoneme sequence “koo: sokudoo: roo: riyoo:” for the first utterance, and “o:” and “o” Since only “o:” is replaced, the phoneme strings are determined to be similar to each other. Therefore, the phoneme sequence “koo: sokudoo: roo: riyoo:” of the first utterance is associated with the command or action corresponding to the phoneme sequence “kousokudourooriyou” of the second utterance. Alternatively, the phoneme sequence “koo: sokudoo: roo: riyoo:” of the first utterance is associated with the phoneme sequence “ko usokudourooriyou”.
In step 211, when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance, the user is inquired whether to associate the voice of the first utterance with the command or action corresponding to the second utterance. For example, it is assumed that the voice command “air conditioner” is registered in the storage unit, the first voice command is “cooler”, and the second voice command is “air conditioner”. The second voice command matches the word or phrase registered in the storage unit, but the phoneme string “kuu: raa:” of the second voice command is similar to the phoneme string “eakon” of the first voice command. Absent. However, since the action performed by the voice command “cooler” and the action performed by the voice command “air conditioner” are common in terms of “turning on the power of the air conditioner”, the user can use the first voice command “cooler”. Can be associated with “turn on the air conditioner” of the air conditioner, which is the action of the second voice command.
Alternatively, in step 211, the speech recognition system (101) selects a command corresponding to the speech of the first utterance when the phoneme sequence of the second utterance is not similar to the phoneme sequence of the first utterance. forgive. The selection is performed, for example, by the user selecting a desired voice command from the list of presented voice commands. When the selection is made, the associating unit registers the voice of the first utterance as a variation of the voice of the second utterance (step 217). If the pronunciation is different, the registration gives the user an opportunity to confirm and register the first voice command as a synonym for the second voice command.
In addition, when registering the voice of the first utterance as a variation of the voice of the second utterance, the registration can be performed according to a criterion for determining whether to register. The criteria can be stored in a policy.
The judgment criteria are as follows.
・ High noise ratio.
-For example, is the S / N ratio higher than a predetermined value? The S / N ratio can vary depending on the spoken environment.
• The frequency of use of the word or phrase you are trying to register.
-Addresses far from the current location will not be used much.
-Only prefectures may be added.
-General nouns and proper nouns-Shrines are common nouns, so they will be used frequently, while proper nouns Kumano Shrine will be used less frequently.
-Well-known proper nouns and not.
-McDonald's (TM) will be used more frequently, but McDonald's Port South Central Point will be used less frequently.
-Frequently used commands and those that are not.
-In a car navigation system, "going home" will be used more frequently, but "today's fortune" will be used less frequently.
-Phoneme sequence-When an impossible phoneme sequence is detected, it is not registered.
-"Nonnn" (3 or more same phonemes follow), but this impossible phoneme sequence varies by language.
・ Use location information and personal preference information.
-The frequency is determined by the distance from the current position and the distance from the home position.
-Registration point -Favorite food, frequent store Further, a step (not shown) is prepared which makes it possible to delete the registered voice data or phoneme string data.

FIG. 2B shows an outline (part 2) of the method for speech recognition according to the embodiment of the present invention.
In step 221, the speech recognition system (FIG. 1A, 101) requests the user to input a third utterance (third utterance) if there is no match in the determination made in step 206. The request can be displayed, for example, on a display device connected to a voice guidance or voice recognition system. The voice or phoneme string of the first utterance is stored in a memory, a hard disk drive, or a solid state disk.
In step 222, the voice recognition system (101) inputs the third utterance by the user through the voice input unit. If the input speech is analog, it may be converted to digital data via an analog-to-digital converter. The recognition engine recognizes a word or phrase for the input utterance and simultaneously generates a phoneme string. The recognized word or phrase for the second utterance and the generated phoneme sequence are stored in memory or a hard disk drive or solid state disk.
In step 223, in response to the input of the third utterance, the voice recognition system (101) determines whether or not the input voice of the third utterance matches the voice registered in the storage unit.
In step 224, if there is a match, the speech recognition system (101) compares the phoneme sequence of the third utterance with the phoneme sequence of the second utterance.
In step 225, the speech recognition system (101) executes an action corresponding to the third utterance if there is a match.
In step 226, the speech recognition system (101) returns to the beginning (A) of the step in FIG. 2B if there is no match.
In step 227, the speech recognition system (101) associates the speech of the second utterance with the command corresponding to the third utterance if the phoneme sequence of the third utterance is similar to the phoneme sequence of the second utterance. In particular, the second voice command is associated with an action corresponding to the third voice command. With this association, by performing the same utterance as the second utterance later, it is possible to execute the action corresponding to the third voice command without requiring the subsequent utterance.
In step 228, the speech recognition system (101) allows a command corresponding to the speech of the first utterance to be selected if the phoneme sequence of the second utterance is not similar to the phoneme sequence of the first utterance. The selection is performed, for example, by the user selecting a desired voice command from the list of presented voice commands. When the selection is made, the associating unit registers the voice of the first utterance as a variation of the voice of the second utterance (step 227).

FIG. 3A shows an outline (part 1) of the mechanism of speech recognition, which is an embodiment of the present invention.
The voice recognition in FIG. 3A indicates voice recognition using a grammar (command).
Command recognition is commonly used in in-vehicle devices.
Assume that the utterance is “atama ga itai”, for example. When the utterance is input, the signal processing unit (301) converts the utterance into an electric voice signal and passes it to the recognition engine (302). The recognition engine (302) recognizes words or phrases in the speech signal using the dictionary (303) and the acoustic model (304).
The dictionary (303) may be a recognition dictionary that is an aggregate of grammars. The grammar is written, for example, “<complaint> = <bodypart> hurts”. For the audio signal, the <bodypart> part is recognized using the dictionary (303). The dictionary contains the head, shoulders, arms, and legs.
The acoustic model (304) uses acoustic features. The acoustic feature represents what frequency characteristic each recognition target phoneme has. As the representation of the acoustic model, a hidden Markov model (HMM) with a mixed normal distribution as an output probability is applicable. The hidden Markov model is applicable because the speech signal can be viewed as a fragmentary or short-time stationary signal.
The recognition engine (302) recognizes that the <bodypart> portion is the “head”. Then, the recognition engine (302) passes the recognized word string “headache” to the application program (305).

FIG. 3B shows an outline (part 2) of the mechanism of speech recognition, which is an embodiment of the present invention.
The speech recognition in FIG. 3B shows speech recognition using large vocabulary recognition (dictation writing).
Assume that the utterance is “posuto-wa-akai-maru”, for example. When the utterance is input, the signal processing unit (311) converts the utterance into an electric voice signal and passes it to the recognition engine (312). The recognition engine (312) recognizes words or phrases for the speech signal using the language model (313) and the acoustic model (314).
In the language model (313), linguistic features are used. A linguistic feature represents a restriction on how phonemes are arranged. For example, immediately after the utterance of “you (anata)”, there is a restriction that the utterance of “ga (ga)” or “ha (wa)” is high. N-gram is used as a representation of the language model. In addition, context-free grammar is used as a language model expression. The n-gram is a model that predicts the next word by looking at the immediately preceding (N-1) words. The context-free grammar is one of the formal grammars in which all generation rules are in the form of V → w. Here, V is a non-terminal symbol, and w is a character string composed of a terminal character and a non-terminal symbol. The term “context-free” means that the non-terminal symbol V can be replaced by w without depending on the context. The n-gram is used, for example, when the language to be recognized is large, and the context-free grammar is used, for example, when the language to be recognized is small enough to be covered manually.
The acoustic model (314) uses acoustic features. The acoustic feature represents what frequency characteristic each recognition target phoneme has. As the representation of the acoustic model, a hidden Markov model (HMM) with a mixed normal distribution as an output probability is applicable. The hidden Markov model is applicable because the speech signal can be viewed as a fragmentary or short-time stationary signal.
The recognition engine (312) recognizes that the audio signal is “post is red circle”. Then, the recognition engine (312) passes the recognized word string “Post is a red circle” to the application program (315).

FIG. 4A shows a flow of processing for speech recognition, which is an embodiment of the present invention.
The processing for performing the speech recognition is an operation example in the Windows (trademark) application of the character user interface (CUI) using IBM EVV.
The main application thread (401) starts processing (steps 402 to 407) in the main application thread of voice recognition processing in response to an enter input from the user.
In step 402, the voice recognition system (101) turns on the microphone in order to start voice recognition by the enter input from the user. When the utterance is started, the utterance is input to the speech recognition system (101).
In step 403, the voice recognition system (101) calls, for example, a voice recognition API esrRecoStartListening to start the processing of the recognition engine. The voice recognition system (101) enters a voice input standby state.
In step 404, the voice recognition system (101) determines that the utterance has been completed by the re-entering input from the user. The voice recognition system (101) turns off the microphone because the voice recognition is finished.
In step 405, the speech recognition system (101) calls, for example, a speech recognition API esrRecoStopListening to end the processing of the recognition engine. The voice recognition system (101) cancels the voice input standby state.
In step 406, the speech recognition system (101) checks the state of the engine through a callback function called from the recognition engine, and waits for acquisition of a speech recognition result.
In step 407, the voice recognition system (101) outputs a voice recognition result. The voice recognition system (101) executes each command when a voice recognition result is output.
Reference numeral 408 in FIG. 4A denotes an ESR recognition engine thread that is sequentially called at regular intervals during the speech recognition processing.
In step 409, when the speech recognition process is started, the speech recognition system (101) calls a callback function to inform the recognition engine of its own state (RECOGNITION STATE) until a speech recognition result is obtained. Keep doing. The callback function is a user-defined function. RECOGNITION STATE is a state that the recognition engine holds internally.
In step 410, the speech recognition system (101) acquires the state of the recognition engine, and confirms whether or not the target state (for example, recognition completion) is obtained in the function. If the determination result is YES, the process proceeds to step 411. On the other hand, if the determination result is NO, the step proceeds to 412.
In step 411, the voice recognition system (101) shares the signal state with the application side.
In step 412, the speech recognition system (101) completes the process and returns to the beginning.
In step 413, the voice recognition system (101) shares the signal state with the application.
In 414 of FIG. 4A, when the speech recognition process is started, the speech recognition system (101) keeps calling the callback function so that the recognition engine notifies its own state until the speech recognition result is obtained. . The callback function is a user-defined function.
In step 415, the speech recognition system (101) inputs a recognition result required by the application. The recognition result is, for example, spelling or ID. The speech recognition system (101) keeps calling the callback function to inform the recognition result.
In step 416, the speech recognition system (101) retrieves desired data in order to hold various information (spell, phoneme sequence, ID, score, etc.) in the form of phrases as speech recognition results.
In step 417, the speech recognition system (101) completes the process and returns to the beginning.
In step 418, the speech recognition system (101) shares the event with the application.

  For example, it is assumed that the words “air conditioner” and “radio” are registered in the recognition dictionary. Suppose the user utters “cooler”. In this case, since the word corresponding to the user's utterance “cooler” does not exist in the recognition dictionary, the recognition result is not applicable. As a recognition engine, when there is no matching result, there is a possibility that a result unintended for the user is output as a result with low probability.

FIG. 4B shows a flow of processing for generating a phoneme string, which is an embodiment of the present invention.
The main application thread (421) starts processing (steps 422 to 427) in the main application thread of phoneme string recognition processing in response to an enter input from the user.
In step 422, the speech recognition system (101) turns on the microphone in order to start phoneme string recognition by the enter input from the user. When the utterance is started, the utterance is input to the speech recognition system (101).
In step 423, the voice recognition system (101) calls, for example, a voice recognition API esrAcbfStartListening to start the processing of the recognition engine. The voice recognition system (101) enters a voice input standby state. Acbf is an abbreviation for phoneme string (Acoustic baseform).
In step 424, the speech recognition system (101) determines that the utterance has been completed by the re-entering input from the user. The voice recognition system (101) turns off the microphone because the voice recognition is finished.
In step 425, the voice recognition system (101) calls, for example, a voice recognition API esrRecoStopListening to end the processing of the recognition engine. The voice recognition system (101) cancels the voice input standby state.
In step 426, the speech recognition system (101) checks the state of the engine through a callback function called from the recognition engine, and waits for acquisition of a phoneme string recognition result.
In step 427, the speech recognition system (101) outputs the recognition result of the phoneme string. The speech recognition system (101) executes each command when the recognition result of the phoneme string is output.
Reference numeral 428 in FIG. 4B denotes an ESR recognition engine thread, which is sequentially called at regular intervals during the phoneme string recognition processing.
In step 429, when the speech recognition process is started, the speech recognition system (101) calls a callback function to notify the recognition engine of its own state (RECOGNITION STATE) until a speech recognition result is obtained. Keep doing. The callback function is a user-defined function. RECOGNITION STATE is a state that the recognition engine holds internally.
In step 430, the speech recognition system (101) acquires the state of the recognition engine, and confirms whether the target state (for example, recognition completion) is obtained in the function. If the determination result is YES, the process proceeds to step 431. On the other hand, if the determination result is NO, the step proceeds to 432.
In step 431, the voice recognition system (101) shares the signal state with the application side.
In step 432, the speech recognition system (101) completes the process and returns to the beginning.
In step 433, the speech recognition system (101) shares the signal state with the application.
In 434 of FIG. 4A, when the phoneme string recognition process is started, the speech recognition system (101) sets a callback function to notify the recognition engine of its own state until a phoneme string recognition result is obtained. Keep calling. The callback function is a user-defined function.
In step 435, the speech recognition system (101) inputs a recognition result required by the application. The recognition result is, for example, a phoneme or ID. The speech recognition system (101) keeps calling the callback function to inform the recognition result.
In step 436, the speech recognition system (101) retrieves desired data in order to hold various information (spell, phoneme string, ID, score, etc.) in the form of phrases as phoneme string recognition results.
In step 437, the speech recognition system (101) completes the process and returns to the beginning.
In step 438, the speech recognition system (101) shares the event with the application.

  For example, assume that the user utters “cooler”. In this case, the “cooler” phoneme string “ku u: ra a:” is generated.

FIG. 5A shows a comparison between phoneme strings, which is an embodiment of the present invention.
In step 501, the speech recognition system (101) compares the first phoneme string with the second phoneme string. The comparison between phoneme sequences cannot be generally described because it depends on the speech recognition system, but for example, as a simple method, the phonemes in the phoneme sequences are compared and the number of coincidence is counted. Alternatively, the phoneme strings are compared using, for example, an API for comparing phoneme strings shown in FIG. 5B.
In step 502, if the first phoneme string and the second phoneme string are similar (YES), the process proceeds to step 503. On the other hand, if the first phoneme string is not similar to the second phoneme string (NO), the process proceeds to step 505.
Whether or not the first phoneme string is similar to the second phoneme string is determined according to the following criteria.
Judgment criteria: Whether the phoneme coincidence in the phoneme string is equal to or lower than a predefined threshold.
In step 503, the voice recognition system (101) determines sound fluctuation. The sound fluctuation information is associated as a change in phoneme sequence for each language, and the information is stored in the storage unit. For example, in the case of Japanese, it is recorded that a phoneme string “ou” is likely to change to a phoneme string “oo:”. The fluctuation between phoneme strings is obtained using, for example, information on the fluctuation of sound prepared for each language. Whether or not the recognized word or phrase is a word that is likely to fluctuate is determined whether or not the phoneme string is partially included in the fluctuation information stored in the storage unit in the given phoneme string. By searching, it is determined.
When it is determined that the phoneme strings are similar in consideration of the fluctuation characteristics of the language, the phoneme strings are registered as variations of other phoneme strings.
If it is determined that the phoneme strings are not similar in consideration of the characteristics of fluctuation by language (NO), the process proceeds to step 505. On the other hand, if it is determined that the phoneme strings are similar in consideration of the fluctuation characteristics of the language (YES), the process proceeds to step 504.
In step 505, the speech recognition system (101) determines that the command corresponding to the first phoneme string is different from the command corresponding to the second phoneme string.
In step 504, the speech recognition system (101) determines that the command corresponding to the first phoneme string is the same as the command corresponding to the second phoneme string. Therefore, the speech recognition system (101) associates the first phoneme string with the command or action corresponding to the second phoneme string. Therefore, the speech recognition system (101) can execute a command corresponding to the second phoneme string by the first phoneme string.

FIG. 5B shows an API for phoneme string comparison, which is an embodiment of the present invention.
API esrCompareBaseforms indicates an example of an API for comparing phoneme strings. The computer system uses this API to determine whether phoneme strings are similar using a threshold value.
API esrBaseformCompareInfo indicates an example of a structure that contains a result of comparison between phoneme strings.

FIG. 6 shows a block diagram of a computer system according to an embodiment of the present invention.
The computer system (601) includes a CPU (602) and a main memory (603), which are connected to a bus (605). The CPU (602) is preferably based on a 32-bit or 64-bit architecture, such as Intel's Xeon (TM) series, Core (TM) series, ATOM (TM) series, Pentium (TM) series, Celeron ™ series, AMD's Phenom ™ series, Athlon ™ series, and the like can be used. A sound board (604) for inputting / outputting audio is connected to the bus (605). A microphone or a speaker is connected to the sound board (604) as necessary. A display (607) such as an LCD monitor is connected to the bus (605) via a display controller (606). The display (607) is used to display information about the software running on the computer system (601) with a suitable graphic interface. A hard disk or silicon disk (609) and a CD-ROM, DVD or Blu-ray drive (610) are also connected to the bus (605) via an IDE or SATA controller (608). The CD-ROM, DVD or Blu-ray drive (610) is used to additionally install programs from the CD-ROM, DVD-ROM or BD to the hard disk or silicon disk (609) as necessary. A keyboard (612) and a mouse (613) are further connected to the bus (605) via a keyboard / mouse controller (611) or via a USB controller (not shown).

  The communication interface (615) conforms to, for example, the Ethernet protocol, and is connected to the bus (605) via the communication controller (614). The communication interface (615) plays a role of physically connecting the computer system (601) and the communication line (616), and is compatible with the TCP / IP communication protocol of the communication function of the operating system of the computer system (601). Providing a network interface layer. The communication line may be a wired LAN environment or a wireless LAN environment based on a wireless LAN connection standard such as IEEE802.11a / b / g / n.

  In addition to the above-described network switch, devices that can be used as a network connection device for connecting hardware such as a computer include a router and a hardware management console. In short, a function capable of returning configuration information such as the IP address and MAC address of a computer connected to an inquiry by a predetermined command from a computer in which a network operation management program is installed. It is. The network switch and router include an ARP table for the address resolution protocol (ARP), which includes a list of IP addresses of computers connected to the router and a corresponding pair of MAC addresses, and can be used to query by a predetermined command. On the other hand, it has a function of returning the contents of the ARP table.

  As described above, the present invention has been described based on the embodiment. However, the content described in the embodiment is an example of the present invention, and those skilled in the art will be able to use various methods without departing from the technical scope of the present invention. It will be clear that variations can be conceived.

1 shows an outline of a speech recognition system that is an embodiment of the present invention. The conceptual diagram of the action with respect to the audio | voice, phoneme, and audio | voice command which is the embodiment of this invention is shown. The outline | summary (the 1) of the method for speech recognition which is the embodiment of this invention is shown. The outline | summary (the 2) of the method for speech recognition which is the embodiment of this invention is shown. The outline | summary (the 1) of the structure of the speech recognition which is the embodiment of this invention is shown. The outline | summary (the 2) of the structure of the speech recognition which is the embodiment of this invention is shown. The flow of the process for speech recognition which is an embodiment of the present invention is shown. The flow of the process for producing | generating the phoneme string which is the embodiment of this invention is shown. The comparison between phoneme strings which is an embodiment of the present invention is shown. An API for phoneme string comparison, which is an embodiment of the present invention, is shown. 1 shows a block diagram of a computer system according to an embodiment of the present invention.

Claims (25)

A computer system for speech recognition of speech input,
In response to the input of the first utterance, a first determination unit that determines whether the input voice of the first utterance matches the voice registered in the storage unit;
A request unit for requesting input of a second utterance when the input voice of the first utterance does not match the voice registered in the storage unit;
A second determination unit that determines whether the voice of the input second utterance matches the voice registered in the storage unit;
A comparison unit that compares the phoneme sequence of the second utterance and the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit;
An association unit for associating the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance ;
A selection unit that allows a command or action corresponding to the second utterance to be selected when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance;
The computer system comprising:   2. The computer system according to claim 1, wherein the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance, based on a matching degree of phonemes in the phoneme sequence.   2. The computer system according to claim 1, wherein when the phoneme string of the second utterance is similar to the phoneme string of the first utterance, whether or not the voices match is determined based on sound fluctuation information. .   The registration unit further includes a registration unit that determines whether to register the phoneme sequence of the first utterance in the storage unit when the phoneme sequence of the second utterance is similar to the phoneme sequence of the first utterance. The computer system described in 1.   The computer system according to claim 4, wherein the registration unit refers to a policy for determining whether to register the phoneme string of the first utterance in the storage unit.   The computer system according to claim 5, wherein the policy is based on at least one of a high noise ratio, a use frequency of words or phrases, and a sequence of phoneme strings.   The computer system according to claim 4, further comprising a deletion unit that determines whether to delete the phoneme string of the registered utterance from the storage unit.   Inquiry unit that inquires of the user whether to associate the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance The computer system of claim 1 further comprising:   9. The computer system according to claim 8, wherein the association unit associates the voice of the first utterance with the command or action corresponding to the second utterance in response to receiving an instruction to perform the association by the user. . The computer system of claim 1 , wherein allowing the selection includes presenting a list of voice commands.   The method of claim 1, wherein associating the voice of the first utterance with a command or action corresponding to the second utterance includes registering the voice of the first utterance as a variation of the voice of the second utterance. Computer system. A second requesting unit for further requesting input of a third utterance when the inputted voice of the second utterance does not match the voice registered in the storage unit;
A third determination unit that determines whether or not the voice of the input third utterance matches the voice registered in the storage unit;
A second comparison unit that compares the phoneme sequence of the third utterance and the phoneme sequence of the second utterance when the speech of the third utterance matches the speech registered in the storage unit;
A second associating unit that associates the voice of the second utterance with the command or action corresponding to the third utterance when the phoneme string of the third utterance is similar to the phoneme string of the second utterance; The computer system according to claim 1.   The computer system according to claim 1, further comprising a first generation unit that generates a first phoneme string corresponding to the first utterance together with the first determination. The computer system according to claim 12 , wherein the first determination unit further determines whether or not the generated first phoneme string matches a phoneme string registered in the storage unit.   The computer system according to claim 1, further comprising a second generation unit that generates a second phoneme string corresponding to the second utterance together with the second determination. The computer system according to claim 14 , wherein the second determination unit further determines whether or not the generated second phoneme string matches a phoneme string registered in the storage unit.   Whether the inputted first utterance voice or the inputted second utterance voice matches the voice registered in the storage unit is determined based on the degree of coincidence of phonemes in the phoneme string, The computer system according to claim 1.   The computer according to claim 1, further comprising: a recording unit that stores the voice of the first utterance in the recording unit when the input voice of the first utterance does not match the voice registered in the storage unit. system.   2. The computer system according to claim 1, further comprising: an execution unit that executes an action corresponding to the first utterance when the input first utterance voice matches the voice registered in the storage unit. .   2. The apparatus according to claim 1, further comprising a second execution unit that executes an action corresponding to the second utterance when the input second utterance voice matches the voice registered in the storage unit. Computer system. A method for speech recognition of speech input, wherein a computer system is
In response to the input of the first utterance, determining whether the input voice of the first utterance matches the voice registered in the storage unit;
Requesting the input of the second utterance when the input voice of the first utterance does not match the voice registered in the storage unit;
Determining whether the voice of the input second utterance matches the voice registered in the storage unit;
Comparing the phoneme sequence of the second utterance and the phoneme sequence of the first utterance when the speech of the second utterance matches the speech registered in the storage unit;
Associating the voice of the first utterance with a command or action corresponding to the second utterance when the phoneme string of the second utterance is similar to the phoneme string of the first utterance ;
When said phoneme strings of the second utterance is not similar to the phoneme string of the first utterance, comprising performing the steps of allowing a selection of a command or action corresponding to the second utterance, the method. The computer system is
Inquiring the user whether to associate the voice of the first utterance with the command or action corresponding to the second utterance when the phoneme string of the second utterance is not similar to the phoneme string of the first utterance
The method of claim 21, further comprising: The computer system is
Associating the voice of the first utterance with a command or action corresponding to the second utterance in response to receiving an instruction to perform the association by the user;
The method of claim 22, further comprising: 24. A method according to any of claims 21 to 23, wherein allowing the selection comprises presenting a list of voice commands. 25. A computer program for speech recognition of speech input, comprising causing a computer system to execute each step of the method according to any one of claims 21 to 24 .

Images