JP2002221987A - Method and device for voice recognition and navigation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a recognition rate by computing noise components more accurately. SOLUTION: Fixed type phrases reproduced from a speaker are collected by a voice recognition microphone prior to the performance of voice recognition of a user. The amount of propagation, that is the difference in comparing voice collected in a step S4 and original sound of the fixed type phrase, is computed as an environmental coefficient Sc. In a step S6, original sound So of a word model in dictionary word data is subtracted from uttered voice Si of the user, the coefficient Sc is subtracted and the result is obtained as a noise template Sn. In a step S7, the template Sn computed in the step S6 is subtracted from the original sound So of the word model in the dictionary word data and the uttered voice Si of the user. The result is compared and the word model for which a highest point is obtained in a voice recognition process is outputted as a recognition result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition method and apparatus and a navigation apparatus using the same.

[0002]

2. Description of the Related Art Conventionally, a voice recognition means of a navigation device is an operation means replacing a remote control device.
The user's voice can be used to set recommended routes to the destination and perform operations such as searching for an arbitrary location. Safety can be improved.

FIG. 4 is a system configuration diagram of a vehicle-mounted navigation device provided with a voice recognition means according to the prior art. In FIG. 4, a voice recognition microphone 21 is for inputting voice from a user. Noise detection microphone 22
Is installed at a predetermined position in the vehicle compartment to input noise inside the vehicle. The voice recognition unit 23 analyzes voice input from the voice recognition microphone 21 and searches for approximate data from dictionary word data to perform voice recognition. The noise detecting means 24 includes the noise detecting microphone 22
Detects vehicle interior noise input from the
Send to 3. The local facility search means 25 searches for a route from the own vehicle position to the destination, facilities around the own vehicle position, and the like. The input / output unit 28 processes input signals from the azimuth sensor 26 and the distance sensor 27, and performs communication and control with peripheral devices by the CPU 33. The DVD-ROM (or CD-ROM) 30 stores map data, voice data, dictionary word data used for voice recognition, and the like. DVD-ROM (or CD-ROM) drive 3
1 reads out, from the DVD-ROM 30, map data, audio data, dictionary word data used for audio recognition, and the like. The communication interface 29 is a DVD-R
It exchanges signals and data between the CPU 33 and an externally connected device such as the OM drive 31. The memory 32 is a memory that stores a plurality of different noise amounts as templates, and stores a noise amount to be added to the word model of the speech recognition dictionary word data stored in the DVD-ROM 30 as a noise template.

When a user's voice is input from the voice recognition microphone 21, the voice recognition means 23
The comparison vocabulary is calculated from the dictionary word data registered in M30.
Selects the amount of noise equivalent to the noise detected in the above from the noise template, compares the data obtained by adding the amount of noise to the comparison vocabulary, and the utterance phrase of the user, and recognizes the closest comparison vocabulary Result. Further, by learning the noise coefficient or the template that has obtained the highest score in speech recognition, the optimum noise amount can be actually added, so that the recognition rate can be increased.

FIG. 5 is a block diagram for explaining a speech recognition method in the prior art. In FIG. 5, dictionary word data D is stored in a DVD-ROM or CD-ROM as a storage means.
A word model for speech recognition recorded in a ROM is recorded. From a noise template T storing a plurality of different noise amounts, a noise coefficient corresponding to noise is determined from the result of the actually detected noise amount. The result of combining the noise amount with the comparison vocabulary in the dictionary word data D by the adding means A and the user's voice input from the microphone M is compared and calculated by the comparing means C so that a matching word model is obtained. decide. The learning means I includes a word model serving as a comparison vocabulary extracted from the dictionary word data D;
The optimum noise amount extracted from the noise template T is synthesized by the adding means A, and the noise coefficient or template with the highest score as a result of comparison with the utterance phrase of the user is learned.

Since the amount of noise to be actually added differs depending on the installation location of the microphone for speech recognition and the microphone for noise detection, the direction of these microphones, the volume of the user's voice, and the like, a template storing a plurality of different amounts of noise addition is used. By preparing in advance, voice recognition corresponding to the noise in the vehicle can be realized.

In the above-mentioned prior art, a noise learning method using a noise detection microphone in addition to a voice recognition microphone has been described. There is also a method of detecting and analyzing a noise component.

[0008]

However, in the above-mentioned conventional speech recognition method, although the user intends to operate the navigation device easily and with ease, the recognition rate is not so high. Had a challenge.
In the case of voice recognition in an in-vehicle device, a vehicle has various causes such as engine noise and tire running noise, and the noise varies in noise component and noise level variously depending on running conditions. External noise may be superimposed on the voice actually uttered by the user, and it may be difficult to compare with the actual voice when comparing the dictionary word data stored in the storage means with a word model which is a vocabulary for comparison. is there.
Also, depending on the installation location and installation direction of the voice recognition microphone, noise components and noise levels differ due to voice reflection and reverberation in the vehicle interior. It is difficult to secure a stable recognition rate due to

Also, a plurality of noise templates are prepared, the corresponding noise is predicted from the vehicle speed or noise collected by a noise detection microphone, and the amount of the noise is added to the original sound of the word model of the dictionary word data. However, when predicting the noise component and the noise level from the vehicle speed, the noise amount is generally large in proportion to the vehicle speed, depending on the transmission switching, the wiper operation, the turn signal operation, and the traveling location. Not necessarily. In addition, since the sound quality and voice level to be spoken differ depending on the user, and the noise component also differs depending on the type of vehicle, it is difficult to separate the volume and sound quality of the speech sound of the actual user from the noise of the vehicle. is there.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and provides a speech recognition method and apparatus capable of more accurately calculating a noise component and improving a recognition rate, and a navigation apparatus using the same. is there.

[0011]

According to the voice recognition method of the present invention, a fixed phrase reproduced before a user's voice recognition is collected by a voice recognition microphone, and the collected voice and the fixed phrase are collected. A propagation attenuation, which is a difference obtained by comparing the original sound with the original sound, is calculated as an environmental coefficient, and then the original sound of the selected comparative vocabulary is subtracted from the dictionary word data based on the voice uttered by the user. , A noise template is created by calculating the amount of noise from the result of subtraction. With this method, the noise component can be calculated more accurately, and the recognition rate can be improved.

Further, in the speech recognition method of the present invention, the original sound of the word model in the dictionary word data is compared with the result obtained by subtracting the calculated noise amount from the uttered voice of the user, and the most approximate result is obtained. Since the word model is used as a recognition result, the noise component can be calculated more accurately, and the recognition rate can be improved.

Further, the voice recognition apparatus of the present invention has a storage means for storing voice recognition dictionary word data, a voice recognition microphone for inputting a user's uttered voice, and a voice guidance speaker for outputting voice guidance. A noise amount is created based on a result obtained by comparing a voice collected by the voice recognition microphone with a fixed phrase reproduced from the speaker and an original sound of the fixed phrase, and the noise amount is stored in the storage unit. Voice recognition means for performing voice recognition by comparing and calculating the original sound of the voice utterance word model. With this configuration, the noise component can be calculated more accurately, and the recognition rate can be improved.

Further, the speech recognition apparatus of the present invention is characterized by learning the calculated noise amount.
Since noise templates corresponding to various different situations can be formed, the recognition rate of the user's uttered voice can be improved.

Further, the present invention is a navigation device provided with the above speech recognition device. When the user operates the voice recognition start button on the remote control or other device, the navigation device plays a voice such as "Please say a voice word" or a fixed phrase that is a signal to prompt the start of speech. The value is determined because it is stored in the storage means of the device, and the fixed phrase is compared with the voice data collected by the voice recognition microphone installed at a specific location to calculate the value. The distance from the guidance speaker can be calculated as the current environmental coefficient. This environmental coefficient varies depending on the sound reflection and reverberation in the cabin depending on the installation position and installation direction of the voice guidance speaker, and also depends on the road conditions and the weather, so the in-vehicle environment immediately before the user's speech is accurately calculated. can do. Further, by learning the calculated environment coefficient, a noise template corresponding to different vehicle types and driving situations can be formed, so that the recognition rate of the user's uttered voice can be improved.

[0016]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an in-vehicle navigation device using a voice recognition method and device according to an embodiment of the present invention. In FIG. 1, a voice recognition microphone 1 is for inputting an uttered voice from a user. The voice recognition means 2 analyzes voice input from the voice recognition microphone 1 and searches for approximate data from dictionary word data.
The voice guidance speaker 3 outputs voices such as route guidance to the destination, travel guidance, and voice guidance for voice recognition. The D / A converter 4 converts and amplifies digital audio data into analog audio data in order to output audio from the speaker 3. The direction sensor 5 detects the traveling direction of the vehicle. The various sensor signals 6 are signals such as a parking switch, a turn signal or a switch input of a transmission. The input / output device 7 processes input signals from the direction sensor 5 and various sensor signals 6, and performs communication and control with peripheral devices by the CPU 12. The DVD-ROM (or CD-ROM) 8 stores map data and voice data, and dictionary word data in which word models as comparison vocabularies used in voice recognition are collected. DVD
The ROM (or CD-ROM) drive 9 is a DVD
Reading map data, voice data, dictionary word data, etc. from the ROM 8; Communication interface 10
Transmits and receives signals and data between the DVD-ROM drive 9 and the CPU 12. The memory 11 stores various data calculated from a plurality of different noise amounts and noise levels and other work data as a template. The CPU 12 controls the entire apparatus, calculates an environmental coefficient, creates a noise template, and the like in speech recognition, and compares the speech model of the user with the word model of the dictionary word data stored in the DVD-ROM 8.

Next, the operation of this embodiment will be described. When a voice recognition request is detected by a user's operation of a remote controller or the like, the CPU 12 converts and amplifies a fixed phrase voice prompting the user to start utterance from the DVD-ROM 8 in which voice data is stored by the D / A converter 4. After doing
The sound is reproduced from the voice guidance speaker 3. The reproduced fixed phrase voice is fed back to the voice recognition microphone 1 and then input.
The propagation attenuation of the sound, which is a difference obtained by performing a comparison operation with the original sound of the fixed phrase sound stored in the DVD-ROM 8, is calculated as an environmental coefficient. By calculating the amount of propagation attenuation of the fixed phrase voice, the distance between the voice guidance speaker 3 and the voice recognition microphone 1 can be calculated. Thereafter, a noise template that is the noise amount is created by calculating the input speech voice of the user, the original sound of the word model in the dictionary word data, and the value calculated as the environmental coefficient. After performing the above processing, the voice recognition means 2 compares the data obtained by adding the noise template to the original sound of the word model in the dictionary word data with the uttered voice of the user, or By comparing data obtained by subtracting the noise template from the uttered voice sound, a more accurate recognition rate can be secured in the current noise environment in the vehicle cabin.

FIG. 2 is an operation flow chart in the present embodiment. Step S1 is an initial setting, where Sc is an environmental coefficient which is a sound propagation attenuation, Sn is a noise template, and Si is a user's uttered voice. In step S2, if there is a request from the user to start speech recognition, in step S3, the navigation device reproduces a fixed phrase such as "Please speak a voice word" and returns the reproduced fixed phrase to the voice recognition microphone 1. Input and capture volume. In step S4, the original sound of the word model of the dictionary word data recorded on the DVD-ROM 8 or the like as the storage means and the sound of the fixed phrase collected in step S3 are calculated, and the difference between the original sound of the word model and the original sound of the actual word model is calculated. Is calculated as the environment coefficient Sc. In step S5, after waiting for the detection of the uttered voice of the user, the original sound So of the word model in the dictionary word data is subtracted from the uttered voice Si of the user in step S6, and the environmental coefficient Sc calculated in step S4 is further subtracted. Then, the result is obtained as a noise template Sn. Step S7 is a comparison process performed by the voice recognizing means 2, which compares the result obtained by subtracting the noise template Sn calculated in step S6 from the original sound So of the word model in the dictionary word data and the user's uttered voice Si, and The word model that obtained the highest score in the recognition process is used as the recognition result as C
Output to PU12. By calculating the environmental coefficient, which is the amount of propagation attenuation of the sound, using a fixed phrase sound whose volume and frequency distribution characteristics are fixed, it is possible to more accurately create a noise template, which is the amount of noise. The recognition rate can be improved.

FIG. 3 is a diagram for explaining an installation position of each device in the vehicle compartment in the present embodiment. In FIG. 3A, D is a driver, M is a voice recognition microphone, and SP is a voice guidance speaker built in the display means. FIG. 3B is similar, but the voice guidance speaker SP can be installed at an arbitrary position. As shown in FIG. 3, the microphone M and the speaker SP are usually arranged at a certain fixed place, and it can be assumed that the microphone M and the speaker SP are at a certain distance from the driver D. There is a difference in the reflection and reverberation of the sound of the sound. As an example, in the case of the speaker SP with built-in display means in FIG. 3A, the propagation of the sound transmitted to the driver D is reduced between the case where the speaker SP is installed toward the driver D and the case where the speaker SP is installed behind the display means. Sorrow is different. Therefore, by making the speech recognition means 2 learn the propagation attenuation amount under various conditions, a noise template corresponding to a different vehicle type and running situation can be formed, and the recognition rate of the user's uttered voice is improved. be able to.

As described above, according to the above embodiment,
Even if a voice recognition microphone or a voice guidance speaker is installed at different positions depending on the vehicle, the noise amount is calculated using a fixed phrase in which the volume and frequency distribution characteristics are determined in advance to create a noise template. Even if the noise environment inside the vehicle differs due to the type of vehicle, driving conditions, weather, operation sounds of vehicle equipment such as wipers, turn signals or audio, etc., the recognition rate for the user's uttered voice is more accurately improved than before. be able to.

Although the above embodiment has been described as an in-vehicle navigation device incorporating a voice recognition device, the present invention can also be configured as a single voice recognition device. Further, in the above-described embodiment, the description has been given as the in-vehicle navigation device. However, the present invention may be configured as a mobile navigation device which is simply carried into the vehicle or detachably mounted on the vehicle.

[0022]

According to the voice recognition method of the present invention, as is apparent from the above embodiment, the fixed phrase reproduced before the user's voice recognition is collected by the voice recognition microphone and collected. The propagation attenuation, which is the difference between the voice and the original sound of the fixed phrase, is calculated as an environmental coefficient, and then the original sound of the comparative vocabulary selected in the dictionary word data is subtracted from the voice uttered by the user, and the environmental coefficient is further calculated. A noise template is created by calculating the amount of noise from the result of subtracting the minutes, and the noise component can be calculated more accurately by using a fixed phrase with a fixed volume and frequency distribution of the reproduced sound. it can. The recognition rate is further improved by comparing the original sound of the word model in the dictionary word data with the result obtained by subtracting the amount of noise from the user's uttered voice, and using the most approximate word model as the recognition result. Can be improved.

Further, the voice recognition apparatus of the present invention has a storage means for storing voice recognition dictionary word data, a voice recognition microphone for inputting a user's uttered voice, and a voice guidance speaker for outputting voice guidance. And a noise amount is created based on the result of comparing the sound collected by the voice recognition microphone with the fixed phrase reproduced from the speaker and the original sound of the fixed phrase, and the noise amount and the voice utterance stored in the storage means are generated. A speech recognition unit for performing speech recognition by comparing and calculating the original sound of the word model is provided. Since the noise component can be calculated more accurately, the recognition rate can be further improved.

Further, the navigation device provided with the voice recognition device of the present invention can be used in a vehicle outside condition such as a running speed and a running place of the vehicle, weather, and a place and an installation direction of a microphone, or a vehicle interior condition such as a volume and a sound quality of a vehicle audio system. Even if a different noise component or noise level is actually generated, just before performing voice recognition, a fixed phrase reproduced by the navigation device itself is input by a voice recognition microphone and compared and calculated.
By calculating the distance between the voice recognition microphone and the voice guidance speaker and using the propagation attenuation of the uttered voice between the user and the voice recognition microphone as the environmental coefficient, the current noise amount is accurately calculated, By learning more environmental factors,
This has the effect of improving the recognition accuracy of voice recognition in various in-vehicle environments. Furthermore, since the amount of noise in the vehicle cabin immediately before the user's speech can be calculated, it is not necessary to monitor and detect the running state of the vehicle using a vehicle speed signal or other sensor signals. Is determined, which has an effect that a more accurate recognition rate can be secured.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a navigation device according to an embodiment of the present invention.

FIG. 2 is a flowchart of a speech recognition process according to the embodiment of the present invention;

FIG. 3 is an explanatory diagram showing an installation position in a vehicle interior according to the embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a conventional navigation device.

FIG. 5 is a block diagram illustrating a conventional speech recognition method.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Voice recognition microphone 2 Voice recognition hand throwing 3 Voice guidance speaker 4 D / A conversion 5 Direction sensor 6 Various sensor signals 7 Input / output means 8 DVD-ROM 9 DVD-ROM drive 10 Communication interface 11 Memory 12 CPU

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI theme coat ゛ (reference) G10L 15/00 G10L 3/00 551Q 15/28

Claims (5)

[Claims] 1. A propagation attenuation amount, which is a difference obtained by collecting a fixed phrase reproduced before a user's voice recognition is performed by a voice recognition microphone and comparing the collected sound with an original sound of the fixed phrase. Is calculated as an environmental coefficient, and then the original sound of the selected comparative vocabulary is subtracted from the dictionary word data based on the voice uttered by the user, and the noise amount is obtained from the result obtained by subtracting the environmental coefficient. A speech recognition method characterized by creating 2. A comparison operation is performed between an original sound of a word model in the dictionary word data and a result obtained by subtracting the calculated amount of noise from the uttered voice of the user, and a word model that is most similar to a recognition result is determined. The voice recognition method according to claim 1, wherein 3. A storage means for storing dictionary word data for voice recognition, a voice recognition microphone for inputting a user's uttered voice, a voice guidance speaker for outputting voice guidance, and a fixed form reproduced from the speaker. A noise amount is created based on a result obtained by comparing the voice collected by the voice recognition microphone with the original sound of the fixed phrase, and the noise amount and the original sound of the voice utterance word model stored in the storage unit are generated. And a voice recognition unit for performing voice recognition by performing a comparison operation with the voice recognition device. 4. The speech recognition apparatus according to claim 3, wherein said calculated noise amount is learned. 5. A navigation device comprising the voice recognition device according to claim 3.