JP2010175921A - Voice recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device capable of reducing a load on operation, and improving precision of voice recognition. <P>SOLUTION: The voice recognition device comprises: a voice input section for inputting voice arising from phrase utterance, to output voice information; a voice recognition section for performing voice recognition processing on the basis of the voice information, to output voice recognition information; an operation input section for inputting the operation performed on the basis of the syllable of the phrase, to output operation information; a storage section for storing a dictionary in which the plurality of phrases are stored; and a selection section which obtains candidate phrases from the dictionary on the basis of the operation information, and which selects the uttered phrase from the candidate phrases on the basis of the voice recognition information. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus.

  As conventional techniques, voice acquisition means for acquiring spoken voice, second input acquisition means for acquiring content input by a second method other than voice, and voice identification for identifying the voice acquired by the voice acquisition means Means, voice recognition dictionary storage means for storing voices and commands in association with each hierarchy, and the hierarchy stored in the voice recognition dictionary storage means by the contents acquired by the second input acquisition means A speech recognition apparatus comprising: a hierarchical identification unit configured to search; and a command content retrieval unit configured to retrieve a voice command identified by the speech identification unit from the hierarchy identified by the hierarchical identification unit. (For example, refer to Patent Document 1).

  According to this voice recognition device, the driver can specify the operation hierarchy only with the sensation and movement of the finger by touchpad operation without turning his / her eyes on the display unit or operation unit even while traveling, and recognition as a target for collation in dictionary data By limiting the vocabulary range, driving safety can be improved, complicated operations can be simplified, operability can be improved, and recognition rate can be improved.

JP 2007-240688 A

  However, in the conventional speech recognition apparatus, the range of the recognition vocabulary is limited only to the vocabulary in the designated operation hierarchy, so that the recognition rate of the vocabulary in the other hierarchy is not improved, which is inconvenient. Met.

  An object of the present invention is to provide a voice recognition device that can reduce the load on operation and further improve the accuracy of voice recognition processing.

  One embodiment of the present invention includes a voice input unit that inputs voice generated by uttering a phrase and outputs voice information, and a voice recognition unit that performs voice recognition processing based on the voice information and outputs voice recognition information An operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information, a storage unit that stores a dictionary in which a plurality of words are stored, and the dictionary based on the operation information A speech recognition apparatus comprising: a selection unit that obtains a candidate word / phrase from the candidate word / phrase and selects the word / phrase uttered from the candidate word / phrase based on the voice recognition information.

  According to another aspect of the present invention, a voice input unit that inputs voice generated by uttering a phrase and outputs voice information, and voice that performs voice recognition processing based on the voice information and outputs voice recognition information A recognition unit; an operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information; a storage unit that stores a dictionary in which a plurality of words are stored; and the voice recognition information And a selection unit that acquires candidate words from the dictionary and selects the words uttered from the candidate words based on the operation information.

  According to another aspect of the present invention, a voice input unit that inputs voice generated by uttering a phrase and outputs voice information, and voice that performs voice recognition processing based on the voice information and outputs voice recognition information A recognition unit; an operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information; a storage unit that stores a dictionary in which a plurality of words are stored; and the voice recognition information The first candidate word / phrase is obtained from the dictionary, the second candidate word / phrase is obtained from the dictionary based on the operation information, and uttered based on the first candidate word / phrase and the second candidate word / phrase. There is provided a voice recognition device comprising a selection unit that selects the word / phrase.

  According to the present invention, it is possible to reduce the load on operations and further improve the accuracy of voice recognition processing.

FIG. 1 is a schematic view of the interior of a vehicle according to the first embodiment of the present invention. FIG. 2 is a block diagram of the speech recognition apparatus according to the first embodiment of the present invention. FIG. 3A is a speech waveform showing a “department store” according to the first embodiment of the present invention, and FIG. 3B is a schematic diagram showing a touch section and a non-touch section. ) Is a schematic diagram illustrating a signal level waveform output from the touch input unit. FIG. 4A is a speech waveform showing “hotel” according to the first embodiment of the present invention, and FIG. 4B is a schematic diagram showing a touch section and a non-touch section, (c ) Is a schematic diagram illustrating a signal level waveform output from the touch input unit. FIGS. 5A to 5C show an example of a touch operation on a word / phrase according to the first embodiment of the present invention. FIG. 6 is a flowchart according to the first embodiment of the present invention. FIG. 7 is a flowchart according to the second embodiment of the present invention. FIG. 8 is a flowchart according to the third embodiment of the present invention.

[First embodiment]
(Vehicle configuration)
FIG. 1 is a schematic view of the interior of a vehicle according to the first embodiment of the present invention. A case will be described in which the voice recognition device of the present invention is mounted on a vehicle and a car navigation device as a controlled device is operated by voice. Moreover, in this Embodiment, the operation by a driver | operator's voice is mainly demonstrated as an operator.

  As shown in FIG. 1, the vehicle 1 is schematically configured to include a steering 10 that is rotatably provided at an end portion of a steering shaft (not shown) of the vehicle 1, a voice recognition device 2, and a car navigation device 3. ing.

  As shown in FIG. 1, the car navigation device 3 includes a display unit 30. For example, a map around the current position of the vehicle 1, an operation menu, and the like are displayed on the display unit 30.

(Configuration of voice recognition device)
FIG. 2 is a block diagram of the speech recognition apparatus according to the first embodiment of the present invention. The speech recognition apparatus 2 limits the candidate words by performing a touch operation on the touch input unit 21 according to the syllables of the phrases, and improves the accuracy of the speech recognition processing. The phrase is, for example, a character string that briefly indicates an operation that can be performed on the controlled device, and may be a word or a sentence.

  As shown in FIGS. 1 and 2, the voice recognition device 2 has a voice input unit 20 provided above a meter panel facing the operator so as to acquire voice and a state in which the operator grasps the steering 10. A touch input unit 21 as an operation input unit provided on the left side of the steering wheel 10, a voice recognition unit 22 to be described later, a template collection 230 to store a template to be described later, and a large number of words and phrases are stored. And a storage unit 23 for storing the dictionary 231 and a selection unit 24 to be described later.

  FIG. 3A is a speech waveform showing a “department store” according to the first embodiment of the present invention, and FIG. 3B is a schematic diagram showing a touch section and a non-touch section. ) Is a schematic diagram showing a signal level waveform output from the touch input unit, and FIG. 4A is an audio waveform showing “hotel” according to the first embodiment of the present invention, ( b) is a schematic diagram showing a touch section and a non-touch section, and (c) is a schematic diagram showing a signal level waveform output from the touch input unit. 3 (a) and 4 (a), the horizontal axis represents time, and the vertical axis represents amplitude. In FIGS. 3 (c) and 4 (c), the horizontal axis represents time, and the vertical axis represents signal level. . The touch section indicates a time during which a touch operation performed on the touch input unit is continued, and the non-touch section indicates a time between the touch section and the touch section, in other words, a time during which no touch operation is performed. ing. Further, “de” and “ho” shown in FIGS. 3B and 4B indicate syllables, and “pau” indicates a silent section.

  The voice input unit 20 includes, for example, a microphone that acquires the voice that is uttered, and outputs a voice waveform shown in FIGS. 3 (a) and 4 (a). This voice waveform is output as voice information.

  The touch input unit 21 includes, for example, a touch sensor that detects a touch operation with a finger, and turns on a change in capacitance due to the touch operation, as illustrated in FIGS. 3C and 4C. And output as an OFF signal level waveform. This signal level waveform is output as operation information. The touch input unit 21 may be, for example, a sensor that can detect a touch operation such as a resistance pressure-sensitive method or an infrared method, or may be a device that can determine ON and OFF, such as a mechanical switch.

  The voice recognition unit 22 performs voice recognition processing on the voice information output from the voice input unit 20, and outputs voice recognition information that is a result of the voice recognition processing to the selection unit 24. The speech recognition information is word / phrase information obtained by performing speech recognition processing.

  The template collection 230 stored in the storage unit 23 is obtained by using speech waveforms such as consonants, vowels, unvoiced plosives (p, t, k), unvoiced friction sounds (s), nasal sounds (n, m, etc.) as templates. It is stored. This template is obtained by quantifying a feature value using MFCC (Mel Frequency Cepstrum Coefficient), for example.

(Operation of voice recognition device)
FIGS. 5A to 5C show an example of a touch operation on a word / phrase according to the first embodiment of the present invention. Hereinafter, the operation of the speech recognition apparatus according to the first embodiment of the present invention will be described in detail according to the flowchart shown in FIG. 6 with reference to the drawings. The case where the operator performs an operation of searching for a department store near the current location by voice will be described below.

  First, in order to search for a department store in the vicinity of the current location, the operator utters “department” as a phrase toward the voice input unit 20, and utters the phrase as shown in FIGS. 3B and 3C. That is, a touch operation is performed on the touch input unit 21 in accordance with the syllable of the phrase (S1).

  Next, the selection unit 24 analyzes the operation information acquired via the touch input unit 21, determines the number of words in the phrase from the number of syllables in the phrase, and becomes a phrase candidate from the dictionary 231 based on the number of characters. Candidate word information is acquired (S2). The candidate phrase information is information on a phrase that matches the number of characters that constitute the phrase.

  Specifically, the selection unit 24 continues for at least a predetermined time (for example, 1 second), an OFF section, that is, the first non-touch section shown in FIG. When the second non-touch period continues again for a predetermined time or more, the number of ONs between the first and second non-touch periods is set as the number of syllables of the phrase. Subsequently, the selection unit 24 determines a prompt sound, a long sound, or the like according to the length of the touch operation in the touch section, and calculates the number of words in the phrase in addition to the number of syllables, for example, with the prompt sound or the long sound as two characters. The selection unit 24 obtains at least one phrase that matches the number of characters from the dictionary 231 as candidate phrase information.

  Here, as shown in FIGS. 5A to 5C, several types of touch operations are conceivable depending on the long sound or the prompt sound included in the phrase. For example, in the case of “pool” shown in FIG. 5A, as shown by a circle in FIG. 5A, a short touch operation is performed in three times, “P”, “−”, “LE”. The case where the touch operation is performed in two times by combining the case of “Pooh”, “Le” and the long touch operation and the short touch operation may be considered. The determination of the long touch operation and the short touch operation by the selection unit 24 may be determined based on, for example, a preset length of the touch section, or may be determined based on the non-touch section. Good and not limited to these.

  In addition, for example, in the case of “Tokyo” shown in FIG. 5B, as indicated by a circle in FIG. 5B, “To”, “U”, “Kyo”, “U” are four times. There are a case where a short touch operation is performed separately and a case where a long touch operation is performed twice, such as “to” and “today”.

  Further, for example, in the case of “Aichi Ken” shown in FIG. 5C, as shown in FIG. 5C, “A”, “I”, “Chi”, “Ke”, “N”, 5 When performing a short touch operation divided into times and when performing a touch operation divided into four times by combining “A”, “I”, “Chi”, “Ken”, short touch operation and long touch operation, etc. Can be considered. “Ken” of “Ken” is also called nasal sound or repellent sound.

  Thus, for example, the selection unit 24 calculates the number of words in a phrase by assuming that one short syllable includes one syllable and one long syllable includes long sound, nose sound, or prompting sound, and that one syllable is two characters. That is, when the long touch operation is performed once and the short touch operation is performed once, the selection unit 24 includes a long syllable, a nasal sound, or a prompt sound from the dictionary 231 and the number of characters is three. Extract words. In addition, the selection unit 24 generates division information for dividing the voice waveform from the touch section and the non-touch section, and transmits the generated information to the voice recognition unit 22.

  Note that, for example, the case where the length of the touch section of “pa” indicating the long sound of “department” and the length of the touch section of any syllable of “hotel” are almost equal. At this time, since the selection unit 24 determines the length of the touch section based on the above-described method, the selection section 24 can accurately determine the length of the touch section in a predetermined word / phrase.

  Next, the selection unit 24 acquires speech recognition information via the speech recognition unit 22, and selects a corresponding phrase from candidate phrases included in the candidate phrase information based on the acquired speech recognition information (S3).

  Specifically, first, the speech recognition unit 22 divides the speech waveform based on the division information acquired from the selection unit 24 as illustrated in FIG. Subsequently, the voice recognition unit 22 performs pattern matching between a voice waveform obtained by dividing the divided section into sections such as vowels and consonants, and a template of a voice waveform included in the template collection 230. The selection unit 24 performs speech recognition processing for each section divided using, for example, an HMM (Hidden Markov Model), and selects words corresponding to the speech information from candidate words based on the result. To do.

  Next, the selection unit 24 transmits the phrase information of the selected phrase to the car navigation device 3 as the controlled device (S4). The phrase information includes information for controlling the car navigation device 3. The car navigation device 3 that has received the phrase information searches for a department store near the current location based on the phrase information and displays the result on the display unit 30. To display.

(Effects of the first embodiment)
According to the speech recognition device in the first exemplary embodiment of the present invention, the following effects can be obtained.
(1) Since the speech waveform can be divided into syllables by a touch operation on the touch input unit, the speech recognition processing is faster and more accurate than the method of dividing the speech waveform using a template. It can be carried out.
(2) Since the number of characters of the recognition target word / phrase is determined by the touch operation, the number of candidate words / phrases can be limited based on the number of characters, and accurate voice recognition processing can be performed.
(3) Since the controlled device can be operated by voice during a driving operation, the operator's line-of-sight movement can be reduced and the load on the operator can be reduced.

[Second Embodiment]
The second embodiment of the present invention will be described below. The second embodiment is different in that a phrase that matches the number of characters calculated based on the touch operation is selected from the candidate phrases selected based on the speech recognition process. In the following embodiments, parts having the same functions and configurations as those of the first embodiment are denoted by the same reference numerals, and description thereof is omitted.

(Configuration of voice recognition device)
The speech recognition apparatus 2 in the present embodiment has the same configuration as the speech recognition apparatus in the first embodiment.

(Operation of voice recognition device)
Hereinafter, the operation of the speech recognition apparatus according to the present embodiment will be described with reference to the flowchart of FIG.

  First, in order to search for a department store in the vicinity of the current location, the operator utters “department” as a phrase toward the voice input unit 20, and utters the phrase as shown in FIGS. 3B and 3C. That is, a touch operation is performed on the touch input unit 21 in accordance with the syllable of the phrase (S20).

  Next, the voice recognition unit 22 acquires voice information via the voice input unit 20 and performs voice recognition processing to generate voice recognition information. The voice recognition unit 22 acquires candidate word / phrase information that is a word / phrase candidate from the dictionary 231 based on the voice recognition information (S21).

  Specifically, as shown in FIGS. 3A and 3B, the speech recognition unit 22 determines a speech waveform indicating a phrase from the first and second non-touch sections. Subsequently, the speech recognition unit 22 performs pattern matching between a speech waveform divided into sections such as vowels and consonants and a template of the speech waveform. For example, the speech recognition unit 22 performs speech recognition processing using an HMM, and acquires candidate phrase information corresponding to the speech information from the dictionary 231 based on the result.

  Next, the selection unit 24 analyzes the operation information acquired via the touch input unit 21 to determine the number of characters constituting the phrase, and selects the phrase from the candidate phrase information corresponding to the speech information based on the number of characters. (S22).

  Next, the selection unit 24 transmits the phrase information of the selected phrase to the car navigation apparatus 3 as a controlled device (S23), and the car navigation apparatus 3 that has received the phrase information receives the phrase information near the current location based on the phrase information. The department store is searched and the result is displayed on the display unit 30.

(Effect of the second embodiment)
According to the speech recognition device in the second exemplary embodiment of the present invention, the number of characters of a phrase uttered by a touch operation is accurately determined, and a phrase corresponding to this number of characters is selected from candidate phrases selected by speech recognition processing. Therefore, accurate voice recognition processing can be performed.

[Third embodiment]
The third embodiment of the present invention will be described below. In the third embodiment, the first candidate word / phrase acquired based on the voice information and the second candidate word / phrase acquired based on the operation information are independently acquired from the dictionary, and the first and second This is different from each embodiment in that a common phrase among the candidate phrases is a spoken phrase.

(Configuration of voice recognition device)
The speech recognition apparatus 2 in the present embodiment has the same configuration as the speech recognition apparatus in the first embodiment.

(Operation of voice recognition device)
Hereinafter, the operation of the speech recognition apparatus according to the present embodiment will be described with reference to the flowchart of FIG.

  First, in order to search for a department store in the vicinity of the current location, the operator utters “department” as a phrase toward the voice input unit 20, and utters the phrase as shown in FIGS. 3B and 3C. That is, a touch operation is performed on the touch input unit 21 in accordance with the syllable of the phrase (S30).

  Next, the voice recognition unit 22 acquires voice information via the voice input unit 20 and performs voice recognition processing to generate voice recognition information. The voice recognition unit 22 acquires first candidate word / phrase information that is a word / phrase candidate from the dictionary 231 based on the voice recognition information. The selection unit 24 analyzes the operation information acquired via the touch input unit 21 to determine the number of characters constituting the phrase, and based on the number of characters, selects second candidate phrase information that is a candidate for the phrase from the dictionary 231. Obtain (S31).

  Next, the selection unit 24 compares the first and second candidate phrases, and selects a phrase common to each candidate phrase (S32).

  Next, the selection unit 24 transmits the phrase information of the selected phrase to the car navigation apparatus 3 as the controlled apparatus (S33), and the car navigation apparatus 3 that has received the phrase information is in the vicinity of the current location based on the phrase information. The department store is searched and the result is displayed on the display unit 30.

(Effect of the third embodiment)
According to the speech recognition device in the third exemplary embodiment of the present invention, the first candidate word / phrase acquired based on the speech recognition information is compared with the second candidate word / phrase acquired based on the operation information. Thus, a voice recognition process with higher accuracy can be performed.

  The speech recognition apparatus of each embodiment described above may prioritize the result obtained by the speech recognition process, for example, when there are a plurality of selectable phrases, or when there are no selectable phrases. Moreover, you may comprise so that it may alert | report via a controlled apparatus etc. so that an input may be prompted again.

  The present invention is not limited to the above-described embodiments, and various modifications and combinations can be made without departing from or changing the technical idea of the present invention.

DESCRIPTION OF SYMBOLS 1 ... Vehicle, 2 ... Voice recognition apparatus, 3 ... Car navigation apparatus, 10 ... Steering, 20 ... Voice input part, 21 ... Touch input part, 22 ... Voice recognition part, 23 ... Memory | storage part, 24 ... Selection part, 30 ... Display unit, 230 ... stationery collection, 231 ... dictionary

Claims (3)

A voice input unit that inputs voice generated by uttering a word and outputs voice information;
A voice recognition unit that performs voice recognition processing based on the voice information and outputs the voice recognition information;
An operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information;
A storage unit for storing a dictionary storing a plurality of words;
A selection unit that acquires candidate words from the dictionary based on the operation information, and selects the words uttered from the candidate words based on the speech recognition information;
A speech recognition device comprising: A voice input unit that inputs voice generated by uttering a word and outputs voice information;
A voice recognition unit that performs voice recognition processing based on the voice information and outputs the voice recognition information;
An operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information;
A storage unit for storing a dictionary storing a plurality of words;
A selection unit that acquires candidate words from the dictionary based on the speech recognition information and selects the words uttered from the candidate words based on the operation information;
A speech recognition device comprising: A voice input unit that inputs voice generated by uttering a word and outputs voice information;
A voice recognition unit that performs voice recognition processing based on the voice information and outputs the voice recognition information;
An operation input unit that inputs an operation performed based on the syllable of the word and outputs operation information;
A storage unit for storing a dictionary storing a plurality of words;
A first candidate word / phrase is obtained from the dictionary based on the speech recognition information, a second candidate word / phrase is obtained from the dictionary based on the operation information, and the first candidate word / phrase and the second candidate word / phrase are obtained. A selection unit for selecting the words uttered based on
A speech recognition device comprising:

Images