JP2006195302A - Speech recognition system and vehicle equipped with the speech recognition system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition system that can enable robust speech recognition even in a state wherein the state of a moving body changes and improve processing efficiency. <P>SOLUTION: The speech recognition system 10 has 1st to 4th grammar dictionaries 62a to 62d and a language dictionary 64 for recognizing speech data inputted from a microphone 20. Those grammar dictionaries 62a to 62d and language dictionary 64 are selectively used based upon a vehicle speed V from a vehicle speed sensor 42, a current position P from a GPS device 44, a yaw rate Y from a yaw rate sensor 46, and a heart rate H from a heart rate sensor 34. Consequently, proper the grammar dictionaries 62a to 62d or the language dictionary 64 matching the state of the vehicle 12 can be selected to improve the processing efficiency of the speech recognition while realizing the speech recognition which is robust. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a voice recognition system to which speaker information input means such as a microphone for inputting information of a speaker at the time of utterance is connected, and a vehicle equipped with the voice recognition system, and more particularly, the state of a moving body changes. The present invention relates to a voice recognition system that performs voice recognition robustly even in a situation in which the vehicle is equipped, and a vehicle including the voice recognition system.

  Recently, many electronic devices have been installed in vehicles, and their functions are becoming more sophisticated. Under such circumstances, in order to facilitate the operation of electronic devices such as a navigation system mounted on a vehicle, a voice recognition system that enables remote operation by voice has been developed.

  Although not limited to vehicles, there is a demand for improving the speech recognition processing efficiency while reliably recognizing speech information spoken by a speaker in a speech recognition system. In order to meet such demands, a speech recognition system has been proposed that selects reference data (speech recognition dictionary) to be searched according to the speaker's utterance time (see, for example, Patent Document 1).

JP 2000-99077 A

  The above speech recognition system is expected to improve the reproducibility of speech recognition if the environment in which the speaker speaks is constant. However, in a moving body such as a vehicle, the surrounding environment for voice recognition changes remarkably according to the travel state and travel location. For this reason, even when the same speaker utters the same word, the speed and clarity of the utterance change. Therefore, in the above speech recognition system, the recognition performance of speech recognition may be lowered depending on the state of the moving body.

  The present invention has been made in view of such problems, and includes a speech recognition system that can perform robust speech recognition and improve processing efficiency even in a situation where the state of a moving body changes, and the speech recognition system. The object is to provide a vehicle.

  A speech recognition system according to the present invention is a speech recognition system in which speaker information input means for inputting speaker information at the time of utterance and state detection means for detecting the state of a moving body are connected, A speech recognition dictionary; input time measuring means for measuring the input time of speaker information input to the speaker information input means; and the state of the moving body detected by the state detecting means and the input time measuring means. Voice recognition dictionary selecting means for selecting the voice recognition dictionary based on the measured input time.

  According to the present invention, the speech recognition dictionary is selected based on the state of the moving body detected by the state detecting unit and the input time of the speaker information measured by the input time measuring unit. For this reason, it is possible to select an appropriate speech recognition dictionary according to the state of the moving body, and it is possible to improve the efficiency of speech recognition while realizing robust speech recognition.

  Here, the speaker information input means is not particularly limited as long as it inputs the information of the speaker at the time of speaking. For example, only the voice information input means such as a microphone for inputting the voice information of the speaker is used. Only for image information input means such as a camera for inputting image information of a speaker such as movement of the mouth of the mouth, so-called lip movement, a combination of voice information input means and image information input means, a combination of a plurality of voice information input means, a plurality of Various configurations such as a combination of the image information input means can be adopted.

  Further, the moving body can be a vehicle, a ship, an amphibious vehicle, a pleasure boat, a helicopter, an airplane, or the like.

  Further, the state of the moving body is the state of the moving body related to voice recognition, that is, detected by a sensor device mounted on the moving body as a state detecting means, such as a vehicle speed sensor, a GPS antenna, a yaw rate sensor, a heart rate sensor, or the like. Refers to the state to be done.

  The speech recognition dictionary includes reference data used for speech recognition. For example, the speech recognition dictionary includes a grammar dictionary including reference data for speech recognition based on recognition based on grammar, and reference data for speech recognition based on free utterance recognition. There are language dictionaries. Note that grammatical recognition is a method of performing speech recognition according to the rules (grammar) required by the system, such as words (eg, “gas station”) and fixed phrases (eg, “turn on radio”). Free speech recognition is a method in which a computer performs speech recognition while interacting (appropriately receiving and answering) a free speech (eg, “Well, is there a good restaurant around here?”).

  The speech recognition system according to the present invention is a speech recognition system in which speaker information input means for inputting speaker information at the time of utterance and state detection means for detecting the state of a moving body are connected, Based on a plurality of speech recognition dictionaries, an input time measuring unit that measures an input time of speaker information input to the speaker information input unit, and a state of the moving body detected by the state detecting unit, Input time correcting means for correcting the input time measured by the input time measuring means; and voice recognition dictionary selecting means for selecting the voice recognition dictionary based on the input time corrected by the input time correcting means. And

  According to the present invention, the input time of the speaker information measured by the input time measuring means is corrected based on the state of the moving body detected by the state detecting means. For this reason, even if the state of the moving body changes, the speech recognition dictionary can be appropriately selected based on the corrected input time. Therefore, the efficiency of voice recognition can be improved while realizing robust voice recognition.

  Here, it is preferable that the voice recognition system switches a voice recognition method according to an input time before or after the correction.

  Thereby, the voice recognition method can be automatically switched without the manual operation of the passenger. For this reason, even if the passenger does not know the voice recognition method, an appropriate voice recognition method can be selected. Further, even when the passenger knows the voice recognition method, the trouble of manually switching the voice recognition method by the passenger can be omitted.

  As the speech recognition method, the above-described grammar recognition, free speech recognition, or the like can be used.

  Further, the correction of the input time measured by the input time measuring means can be performed by converting the state of the moving body into a coefficient and multiplying the input time by this coefficient.

  Thereby, even if the state of the moving body changes, it is possible to correct the voice input time so as to complement this change. Therefore, it is sufficient to provide a speech recognition dictionary in the reference state of the moving body for each speech input time, and the speech recognition dictionary can be selected efficiently.

  The vehicle according to the present invention includes any one of the above-described voice recognition systems.

  According to the present invention, the speech recognition dictionary is selected based on the state of the moving body detected by the state detecting unit and the input time of the speaker information measured by the input time measuring unit. For this reason, it is possible to select an appropriate speech recognition dictionary according to the state of the moving body, and the effect of improving the speech recognition processing efficiency while achieving robust speech recognition is achieved.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of a voice recognition system according to the present invention and a vehicle including the voice recognition system will be described with reference to FIGS.

  As shown in FIG. 1, a speech recognition system 10 according to an embodiment of the present invention is mounted on a vehicle 12 (moving body).

  The voice recognition system 10 is connected to a microphone (speaker information input means) 20 for inputting the voice of a passenger (speaker) 14 via an in-vehicle communication network 18. The microphone 20 is provided in the vicinity of the boundary between the roof and the windshield glass in the vehicle. The microphone 20 may be provided in places other than this (for example, a roof lower surface part, an instrument panel part, a headrest, a passenger's shoulder part, a headset, etc.).

  A navigation system 28, an engine controller 30, a panel operation unit 32, a heart rate sensor 34 and a sweat sensor 36 are connected to the in-vehicle communication network 18, and these devices can communicate data with each other via the in-vehicle communication network 18. It is. As will be described later, the heart rate sensor 34 and the sweat sensor 36 can be used as the state detection means of the present invention.

  A GPS (Global Positioning System) device 44 is connected to the navigation system 28, and coordinate information of the current position P of the vehicle 12 can be acquired.

  The engine controller 30 controls the engine 40 and can detect the vehicle speed V of the vehicle 12 by the vehicle speed sensor 42. Further, the engine controller 30 can acquire the yaw rate Y of the vehicle 12 (scale adjusted to be in the range of 0-100) by the yaw rate sensor 46.

  The panel operation unit 32 is provided with switches and buttons operated by the passenger 14, and the operation control of the navigation system 28 can be performed from here.

  As shown in FIG. 2, the heart rate sensor 34 and the sweat sensor 36 are attached to the steering 22 of the vehicle 12. The heart rate sensor 34 measures the heart rate of the passenger 14 through the hand of the passenger 14 and outputs the measurement result to the voice recognition system 10 as a digital signal. The sweat sensor 36 measures the amount of sweat in the hand of the occupant 14 and outputs the measurement result to the voice recognition system 10 as a digital signal.

  The heart rate sensor 34 and the sweat sensor 36 only need to be able to measure the heart rate and the amount of sweat of the occupant 14, and their arrangement and performance can be changed according to design conditions.

  As shown in FIG. 3, the speech recognition system 10 includes an A / D conversion unit 50, a temporary storage unit 52, a feature extraction unit 54, a speech pattern recognition unit 56, an acoustic dictionary 58, and a speech recognition dictionary storage unit. 60.

  The A / D conversion unit 50 converts an analog audio signal of the passenger 14 input from the microphone 20 into digital audio data. Specifically, an analog audio signal is converted into digital audio data using a quantization technique such as PCM (pulse code modulation).

  The temporary storage unit 52 temporarily stores the digital audio data converted by the A / D conversion unit 50 as temporary storage data. As the temporary storage unit 52, for example, a commercially available memory can be used.

  The feature extraction unit 54 extracts a characteristic part from the temporarily stored data temporarily stored in the temporary storage unit 52 as feature extraction data. For example, the temporally stored data is subjected to frequency analysis using fast Fourier transform (FFT) to extract feature extraction data.

  The voice pattern recognition unit 56 recognizes a voice pattern from the feature extraction data extracted by the feature extraction unit 54 using the voice recognition dictionary and the acoustic dictionary 58 in the voice recognition dictionary storage unit 60. As the speech pattern recognition method, for example, a pattern matching method or a statistical method can be used. As a statistical technique, a hidden Markov model having a probabilistic finite state can be cited. In the hidden Markov model, the speech can be recognized with high probability by learning the speech model.

  The acoustic dictionary 58 includes acoustic features for each small unit (phoneme) of human speech as a plurality of reference acoustic patterns. The voice pattern recognition unit 56 calculates the acoustic feature of the feature extraction data using this reference acoustic pattern.

  The speech recognition dictionary storage unit 60 includes first to fourth grammar dictionaries 62a to 62d that include reference data for speech recognition based on recognition based on grammar, and a language dictionary 64 that includes reference data for speech recognition based on free speech recognition. Have

  As shown in FIG. 4, the first to fourth grammar dictionaries 62 a to 62 b are provided with reference data for speech recognition by grammar according to the voice input time of the passenger 14. For example, a voice pattern for each word or fixed phrase is stored as reference data according to the pronunciation time of each reference data.

  The language dictionary 64 includes reference data for free utterance recognition. For example, the above-described grammar dictionaries or interjections (e.g., "Ut" and "Um") that have many speech patterns for each word or fixed sentence as speech recognition data, such as the first to fourth grammar dictionaries 62a to 62d. The speech recognition dictionary (rejection dictionary) or the like having a speech pattern unnecessary for speech recognition as speech recognition data can be configured. Note that the number of language dictionaries 64 is not limited to one, and a plurality of language dictionaries 64 may be provided according to the state of the vehicle 12 and the like.

  Returning to FIG. 3, the speech recognition system 10 further includes an input time measuring unit 66 and a speech recognition control unit 68.

  The input time measuring unit 66 measures the voice input time of the passenger 14 based on the temporarily stored data temporarily stored in the temporary storage unit 52. Specifically, a predetermined threshold value related to the volume is set, and the time during which data exceeding the threshold value continues is measured as the voice input time.

  The voice recognition control unit 68 issues various commands to the voice pattern recognition unit 56, the input time measurement unit 66, and each sensor device (vehicle speed sensor 42, GPS device 44, yaw rate sensor 46, heart rate sensor 34, etc.). Voice recognition is performed with this. The function of the voice recognition control unit 68 will be described later.

  Next, a procedure for recognizing the voice of the passenger 14 using the voice recognition system 10 configured as described above will be described.

  In step S <b> 1 of FIG. 5, when the voice recognition control unit 68 detects a voice recognition start input from the passenger 14 to the navigation system 28, the voice recognition control unit 68 turns on the microphone 20. Next, when the passenger 14 utters a sound, the microphone 20 that has detected the sound outputs an analog sound signal corresponding to the sound. The analog audio signal is converted into digital audio data by the A / D conversion unit 50 and then temporarily stored in the temporary storage unit 52.

  Next, in step S <b> 2, the voice recognition control unit 68 causes the input time measurement unit 66 to measure the input time of the temporarily stored data temporarily stored in the temporary storage unit 52. The input time (input time T1 before correction) measured by the input time measurement unit 66 is output to the voice recognition control unit 68 as digital data.

  In step S3, the voice recognition control unit 68 reads signals from various sensor devices (vehicle speed sensor 42, GPS device 44, yaw rate sensor 46, heart rate sensor 34). That is, the vehicle speed sensor 42 is the vehicle speed V, the GPS device 44 is the current position P, the yaw rate sensor 46 is the yaw rate Y (scale adjusted to be in the range of 0-100), and the heart rate sensor 34 is the heart rate. Read H. Based on the read value, the input time T1 before correction measured by the input time measuring unit 66 is corrected to calculate the input time T2 after correction. That is, the corrected input time T2 is calculated using the following equation (1).

T2 = T1 × C _v × C _P × C _Y × C _H (1)

Here, C _v is the time correction factor of the vehicle speed V, C _P is the time correction factor of the current position P, C _Y the time correction factor of the yaw rate Y, C _H is the time correction factor in heart rate H. Each correction coefficient C _v , C _P , C _Y , C _H is obtained based on the relationship shown in FIG.

  For example, when the input time before correction T1 = 0.9 (seconds), the vehicle speed V = 62 (km / h), the current position P = mountain portion, the yaw rate Y = 20, and the heart rate H = standard, the following formula (2 ), The corrected input time T2≈1.04 (seconds).

T1 C _v C _P C _Y C _H
T2 = 0.9 × 1.05 × 1.1 × 1.0 × 1.0≈1.04 (2)

  In FIG. 6, the correction coefficient increases as the vehicle speed V increases. This is because the higher the vehicle speed V, the higher the tension of the passenger 14 and the higher the speaking speed of the passenger 14.

  The same applies to the current position P, the yaw rate Y, and the heart rate H, and the correction coefficient is set to increase as the tension state of the passenger 14 increases.

  In addition, regarding the current position P in FIG. 6, determination of urban areas, mountainous areas, and highways is stored in advance in the navigation system 28 and positional information (latitude, longitude, etc.) included in the signal received by the GPS device 44 from the satellite. This is done by comparing / determining the map information.

  Further, for the heart rate H, a standard value is set in advance in the voice recognition control unit 68. Then, the voice recognition control unit 68 compares the heart rate H from the heart rate sensor 34 with this standard value, and determines a correction method.

  In step S4, the speech recognition control unit 68 determines a recognition method (recognition dictionary) based on the corrected input time T2 calculated in step S3. That is, as shown in FIG. 7, when the corrected input time T2 is less than 2 seconds, grammatical recognition is performed using one of the first to fourth grammar dictionaries 62a to 62d. When the corrected input time T2 is 2 seconds or longer, free speech recognition is performed using the language dictionary 64. In the above example, the post-correction input time T2 is less than 2 seconds since T2≈1.04 seconds. For this reason, recognition by grammar is selected as the recognition method. As the speech recognition dictionary, the third grammar dictionary 62c corresponding to the corrected input time T2 is selected from the first to fourth grammar dictionaries 62a to 62d (see FIG. 4).

  In step S5, the speech recognition control unit 68 notifies the speech pattern recognition unit 56 of the selected speech recognition dictionary (the third grammar dictionary 62c in the above example).

  In step S <b> 6, the voice pattern recognition unit 56 performs voice recognition of the feature extraction data extracted by the feature extraction unit 54 using the third grammar dictionary 62 c notified from the voice recognition control unit 68. The recognition operation here uses the above-described recognition method such as a pattern matching method or a statistical method.

  In step S <b> 7, the voice pattern recognition unit 56 outputs the recognition result to the voice recognition control unit 68. The voice recognition control unit 68 outputs the recognition result input from the voice pattern recognition unit 56 to the navigation system 28 via the in-vehicle communication network 18 and ends the voice recognition process. The navigation system 28 performs navigation based on the input recognition result.

  As described above, in the speech recognition system 10 according to this embodiment, the speech recognition dictionary used for recognition of speech input from the microphone 20 is determined based on the state of the vehicle 12 (the vehicle speed V from the engine controller 30 and the yaw rate sensor 46). Selected based on the yaw rate Y). For this reason, it is possible to select an appropriate speech recognition dictionary corresponding to the state of the vehicle 12, and it is possible to improve the efficiency of speech recognition while realizing robust speech recognition.

  In the voice recognition system 10, the voice input time (pre-correction input time T <b> 1) is corrected based on the state of the vehicle 12. For this reason, even if the state of the vehicle 12 changes, the speech recognition dictionary can be appropriately selected based on the corrected speech input time (corrected input time T2) and the state of the vehicle 12. Therefore, the efficiency of voice recognition can be improved while realizing robust voice recognition.

  Furthermore, the speech recognition system 10 selects grammatical recognition or free utterance recognition according to the speech input time (corrected input time T2 in this embodiment). For this reason, even if the passenger 14 does not know the voice recognition method, an appropriate voice recognition method can be selected. Further, even when the passenger knows the voice recognition method, the time and effort of the passenger can be omitted.

  In addition, the voice recognition system 10 corrects the voice input time by converting the state of the vehicle 12 into a coefficient and multiplying this coefficient by the voice input time (pre-correction input time T1). For this reason, even if the state of the vehicle 12 changes, the voice input time can be corrected so as to complement this change. Therefore, a speech recognition dictionary in the reference state of the vehicle 12 may be provided for each speech input time, and the speech recognition dictionary can be selected efficiently.

  Note that the present invention is not limited to the above-described embodiment, and it is needless to say that various configurations can be adopted based on the description in this specification.

  For example, in the above-described embodiment, the vehicle 12 is exemplified as the moving body. However, the moving body is not limited thereto, and may be a moving body such as a ship, an amphibious vehicle, a pleasure boat, a helicopter, and an airplane.

  Moreover, in the said embodiment, although the state of the vehicle 12 was used for correction | amendment of voice input time, if the speech recognition dictionary is selected based on the state of the vehicle 12 and voice input time, it will not be restricted to this. For example, a correspondence table between the combination of the state of the vehicle 12 and the voice input time and the voice recognition dictionary may be created and stored in advance, and the voice recognition dictionary may be selected using the correspondence table. In addition, the voice recognition dictionary may be set as a default in relation to the voice input time, and may be shifted to a voice recognition dictionary with higher recognition performance or a voice recognition dictionary with higher processing efficiency depending on the state of the vehicle 12. it can.

  In the above embodiment, the vehicle speed V, the current position P, the yaw rate Y, and the heart rate H are used as the state of the vehicle 12, but this is not necessarily limited to this combination. For example, it is possible to use only the vehicle speed V. Further, the state of the vehicle 12 other than the above, for example, the average frequency F, the sweating amount S, and the like can be used. The average frequency F can be calculated based on values such as the vehicle speed V and the engine speed. The perspiration amount S can be detected using a perspiration sensor 36.

  The microphone 20 is used as the speaker information input means. However, the present invention is not limited to this as long as it can acquire voice recognition information such as an image acquisition means such as a CCD camera that images the movement of the mouth of the passenger 14, so-called lip movement. I can't. Also, a plurality of speaker information input means can be used.

  Although the first to fourth grammar dictionaries 62a to 62d and the language dictionary 64 are used as the voice recognition dictionary, it is sufficient that there are at least two voice recognition dictionaries. Further, only one of a grammar dictionary and a language dictionary may be used, and a speech recognition dictionary based on another speech recognition method may be used.

  Furthermore, in the above embodiment, the acoustic dictionary 58 relating to the acoustic features for each phoneme was not included in the selection target speech recognition dictionary. However, by providing a plurality of acoustic dictionaries 58, a selection target speech recognition dictionary may be used. it can. For example, it is possible to provide a plurality of acoustic dictionaries 58 for each state of the vehicle 12 and the sound generation time of the reference sound pattern, and to select them based on the state of the vehicle 12 and the input time of phonemes.

  The speech recognition method is switched according to the input time after correction T2, but can be switched according to the input time before correction T1.

  Although the method of multiplying the voice input time by a predetermined coefficient is employed, for example, a method of adding / subtracting according to the state of the vehicle 12 can be employed.

  The above embodiment is a system that is completed inside the vehicle. However, the present invention is not limited to this. For example, the voice pattern recognition unit 56 and the voice recognition dictionary storage unit 60 are arranged in an information center outside the vehicle, and voice communication is performed using wireless communication. It is also possible to perform recognition.

1 is a block configuration diagram of a vehicle equipped with a voice recognition system according to an embodiment of the present invention. It is a top view of the steering of the vehicle. It is a block block diagram of the said speech recognition system. It is a figure which shows the relationship between the grammar dictionary in the said speech recognition system, and a passenger | crew's voice input time. It is a flowchart which shows the procedure of the speech recognition process in the said speech recognition system. It is a figure which shows the relationship between the value of each sensor apparatus in the said voice recognition system, and the correction | amendment method of the said voice input time. It is a figure which shows the relationship between the said voice input time in the said speech recognition system, a speech recognition method, and a speech recognition dictionary.

Explanation of symbols

10 ... Voice recognition system 12 ... Vehicle (moving body)
14 ... Passenger (speaker) 20 ... Microphone (speaker information input means)
34 ... Heart rate sensor (state detection means) 42 ... Vehicle speed sensor (state detection means)
44 ... GPS device (state detection means)
46 ... Yaw rate sensor (state detection means)
62a-62d ... Grammar dictionary (voice recognition dictionary)
64 ... Language dictionary (voice recognition dictionary)
66 ... Input time measuring unit (input time measuring means)
68. Voice recognition control unit (input time correcting means and voice recognition dictionary selecting means)

Claims (4)

A speech recognition system in which speaker information input means for inputting speaker information at the time of utterance and state detection means for detecting the state of a moving body are connected,
Multiple speech recognition dictionaries,
Input time measuring means for measuring the input time of speaker information input to the speaker information input means;
A speech recognition dictionary selecting means for selecting the speech recognition dictionary based on the state of the moving body detected by the state detecting means and the input time measured by the input time measuring means. . A speech recognition system in which speaker information input means for inputting speaker information at the time of utterance and state detection means for detecting the state of a moving body are connected,
Multiple speech recognition dictionaries,
Input time measuring means for measuring the input time of speaker information input to the speaker information input means;
An input time correcting means for correcting the input time measured by the input time measuring means based on the state of the moving body detected by the state detecting means;
A speech recognition system comprising: speech recognition dictionary selecting means for selecting the speech recognition dictionary based on the input time corrected by the input time correcting means. The voice recognition system according to claim 1, wherein the moving body is a vehicle. A vehicle comprising the voice recognition system according to claim 3.

Images