JP2002196786A - Speech recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition device with high recognition rate. SOLUTION: In the speech recognition device which recognizes the voice of a user, the device has a learning means to perform learning on the basis of the voice input by the user, and a dictionary to be corrected for the specific speaker based on the learned result by the learning means, the learning means collates the inputted voice with the dictionary, and the learned result is returned to the initial value side on the basis of this collated result only by the prescribed amount.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle-mounted device, and more particularly to a learning function of the device.

[0002]

2. Description of the Related Art In recent years, various devices are mounted on a vehicle as typified by a car navigation system, and a voice recognition device is used as an input device for a driver during driving to operate the devices safely. It has been developed.

[0003] As such a conventional speech recognition apparatus, for example, an apparatus disclosed in Japanese Patent Application Laid-Open No. H6-130985 is known. Conventional speech recognition devices used for specific speakers input a user's voice in order to improve the recognition rate by modifying the dictionary prepared for unspecified speakers for specific speakers. Need to learn.

[0004] For example, FIG. 1 is a block diagram showing a configuration of a general speech recognition apparatus, and includes a dictionary 1, a learning unit 2, a recognition unit 3, and a learning recognition control unit 4. In such a general speech recognition device, first, the learning recognition control unit 4 is set to a learning mode, and a user sequentially inputs a plurality of predetermined words, for example, and the learning recognition control unit 4 outputs the words. Learning part 2
To send to. Thereby, the learning unit 2 sequentially corrects the dictionary 1 for a specific speaker. After the learning is completed, the learning recognition control unit 4 is set to the recognition mode. Then, the unknown input voice is sent to the recognition unit 3, and the recognition unit 3 outputs the recognition result by referring to the dictionary corrected for the specific speaker.

There is also known a speech recognition apparatus that performs learning in which the dictionary is constantly modified little by little by the same learning during normal use as well as in the learning mode.

[0006]

The above-mentioned conventional speech recognition apparatus, after learning for a specific speaker is completed,
Since the input speech is recognized by referring to the dictionary modified for the specific speaker, the recognition rate improves if the same speech style is used when learning in the learning mode, but on the contrary, the speech style changes a little. Therefore, there is a problem that the recognition rate of the input voice is reduced because the dictionary is different from the dictionary to be referred to.

The present invention has been made to solve such a problem, and has as its object to provide a speech recognition device having a high recognition rate.

[0008]

A speech recognition apparatus according to the present invention is a speech recognition apparatus for recognizing a user's voice, wherein learning means for learning based on a user's voice input;
A dictionary that is modified for a specific speaker based on the learning result of the learning means; the learning means collates the input voice with the dictionary; and initializes the learning result by a predetermined amount based on the collation result. It returns to the value side.

The learning means compares the input voice with the dictionary and the dictionary with the input voice and the number of matches is equal to or less than a predetermined value. To return to the side.

The learning means changes the predetermined amount for returning the learning result to the initial value side in accordance with the ratio between the number of times of voice input and the number of times of matching.

Further, the learning means calculates a learning progress degree based on the learning result and displays the learning progress degree.

Further, the learning means gives an alarm when the degree of progress of the learning falls below a predetermined value.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 Hereinafter, an embodiment of the present invention will be described. The configuration is the same as that of the above-described general speech recognition device, and a detailed description is omitted. The operation of the first embodiment will be described with reference to the flowchart of FIG. Further, the learning process of the speech recognition device is generally known, and a detailed description thereof will be omitted.

Here, it is assumed that the process of FIG. 2 is executed at regular intervals. First, 1 is added to the counter t in step S11. The counter t counts the number of executions of this process, that is, the elapsed time.
In step S12, the presence or absence of an input voice is determined. If there is no input voice, the process proceeds to step S17 described later. On the other hand, if there is an input voice, the process proceeds to step S13, where the contents of the dictionary 1 corrected for the specific speaker based on the learning result already obtained by the learning process are compared with the input voice, and the process proceeds to step S14. To determine whether they match. Here, the match may be determined based on whether or not the input voice and the contents of the dictionary completely match, or may be determined to match if the degree of matching between the two is equal to or greater than a predetermined value.

If it is determined in step S14 that they match, one is added to the counter C in step S15, and if they do not match, the counter C is added in step S16.
Is subtracted by 1. Steps S14, S15, S16
Is to count the number of times the input voice matches the contents of the dictionary.

Next, in step S17, it is determined whether or not the counter t is equal to or more than a predetermined value, that is, whether or not a predetermined time has elapsed. If not, the process is terminated. Proceed to S18. In step S18, it is determined whether or not the counter C is equal to or less than a predetermined value. The process proceeds to step S19, in which a process of returning the learning result to the initial value side by a predetermined amount is executed. On the other hand, step S18
If the counter C is equal to or more than the predetermined value, the number of times that the contents of the dictionary match the input voice is large, and it is determined that the user's speaking style is constant, and the process of returning the learning result to the initial side is not executed. In any case, the process finally proceeds to step S20, where the counter t and the counter C are reset and the process ends.

By executing the above-mentioned processing, the learning result returns to the initial value side unless a voice that matches the contents of the dictionary is input a predetermined number of times or more within a predetermined time. Since the consciousness of the user works to prevent the learning result from being erased and the voice is input in a consciously constant manner of speech, the input rate of the voice recognition device is constant, so that the recognition rate is improved.

Embodiment 2 FIG. Next, another embodiment of the present invention will be described. In the first embodiment, the learning result returns to the initial value side unless a voice matching the contents of the dictionary is input a predetermined number of times or more within a predetermined time. Similarly, the learning result returns to the initial value side. This can be prevented by performing the process of determining the number of times of inputting the voice that matches the contents of the dictionary for each predetermined number of inputs and returning the learning result to the initial value side.

More specifically, as shown in the flowchart of FIG. 3, first, in step S21, an input voice is awaited. If there is an input voice, the process proceeds to step S22, and 1 is added to the counter n. This counter n counts the number of times that a voice is input. Next, the process proceeds to step S23, and in step S24, it is determined whether or not both match. Here, the match may be determined based on whether or not the input speech and the contents of the dictionary completely match, as in the first embodiment. May be determined.

If it is determined in step S24 that they match, one is added to the counter C in step S25, and if they do not match, the counter C is added in step S26.
Is subtracted by 1. Steps S24, S25, S26
Is to count the number of times the input voice matches the contents of the dictionary.

Next, in step S27, it is determined whether or not the counter n is equal to or more than a predetermined value, that is, whether or not the voice has been input a predetermined number of times.
The process returns to 1 and waits for the next input voice. On the other hand, if the number has been input more than the predetermined number, the process proceeds to step S28. In step S28, it is determined whether or not the value of the counter C is equal to or less than a predetermined value. The process proceeds to step S29, in which a process of returning the learning result to the initial value side by a predetermined amount is executed. on the other hand,
If the counter C is equal to or more than the predetermined value in step S28,
The process of returning the learning result to the initial side by judging that the contents of the dictionary match the input voice many times and that the user's speech style is constant is not executed. In any case, the last step S
Proceeding to 30, the counter n and counter C are reset, and the process returns to step 21 to wait for the next input voice.

By executing the above-described processing, when a voice is actually input a predetermined number of times, if a voice that matches the contents of the dictionary is not input a predetermined number of times or more, the learning result is reduced to the initial value. Because the user is conscious of trying to prevent the learning result from being erased and inputs speech in a consciously constant manner of speech, the input speech is used as a speech recognition device. Since it is constant, the recognition rate is improved. Further, it is possible to prevent the learning result from returning to the initial value side when the non-input state continues for a long time, and the number of mismatches between the actually input voice and the dictionary, that is, the user's The learning content can be returned to the initial value side according to the variation in the way of speaking.

In each of Embodiments 1 and 2 described above, when the user's way of speaking varies, the processing is performed so that the learning result is returned to the initial value side by a predetermined amount. It may be changed according to the degree of variation. That is, in the second embodiment, the learning result is returned to the initial value side by changing the predetermined amount according to the ratio of the number n of times the voice is input and the number C of times when the input voice matches the contents of the dictionary. It may be. by this,
The consciousness of the user who tries to keep the speech style constant works more strongly, and the recognition rate can be further improved.

The degree of progress of learning (degree of retreat) may be displayed on an instrument panel or a car navigation monitor. As a result, the user can confirm that the learning result returns to the initial value side, and the user's consciousness for stabilizing the way of speaking works more strongly, and the recognition rate can be further improved.

Furthermore, when the degree of progress of learning (degree of retreat) becomes equal to or less than a predetermined value, a warning may be issued to the user. Thereby, when the learning result has returned to the initial value to some extent, it is possible to make the user recognize that the learning process is necessary again, and without continuing to use the device while the recognition rate is lowered, The learning process can be performed again to improve the recognition rate before use.

[0026]

The speech recognition apparatus according to the present invention is a speech recognition apparatus for recognizing a user's voice, comprising: learning means for learning based on a user's voice input; and a specific speech based on a learning result by the learning means. The learning means compares input speech with the dictionary, and returns the learning result to the initial value side by a predetermined amount based on the result of the comparison. Will return to the initial value side, and the user will work to prevent the learning result from being erased, and will input speech in a consciously constant manner of speech. Since the input voice is constant, the recognition rate is improved.

Further, the learning means compares the input voice with the dictionary and compares the input voice with the dictionary and determines whether the number of matches between the voice and the dictionary is equal to or less than a predetermined value. When the voice is actually input a predetermined number of times, if the voice that matches the contents of the dictionary has not been input more than a predetermined number of times, the learning result will return to the initial value side Therefore, the user works to prevent the learning result from being erased, and inputs the voice in a consciously constant manner of speech.As a speech recognition device, the input voice is constant and the recognition rate is improved. I do. Further, it is possible to prevent the learning result from returning to the initial value side when the non-input state continues for a long time, and the number of mismatches between the actually input voice and the dictionary, that is, the user's The learning content can be returned to the initial value side according to the variation in the way of speaking.

The learning means changes the predetermined amount for returning the learning result to the initial value side in accordance with the ratio between the number of times of voice input and the number of times of matching. It works more strongly and can improve the recognition rate.

Further, the learning means calculates a learning progress degree based on the learning result and displays the learning progress degree. The user can confirm that the learning result returns to the initial value side. The consciousness of the user who tries to keep the speech style constant works more strongly, and the recognition rate can be further improved.

Further, the learning means gives an alarm when the learning progress degree falls below a predetermined value. When the learning result has returned to the initial value side to some extent, the user is asked again. It is possible to recognize that the learning process is necessary, and it is possible to perform the learning process again without continuing to use the device while the recognition rate is lowered, and to use the device with an improved recognition rate.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a general voice recognition device.

FIG. 2 is a flowchart showing the operation of the first embodiment.

FIG. 3 is a flowchart showing an operation of the second embodiment.

[Explanation of symbols]

 Reference Signs List 1 dictionary 2 learning unit 3 recognition unit 4 learning recognition control unit

Claims (5)

[Claims] 1. A speech recognition apparatus for recognizing a user's voice, comprising: a learning unit for learning based on a user's voice input; and a dictionary modified for a specific speaker based on a learning result by the learning unit. A speech recognition apparatus, wherein the learning unit collates the input speech with the dictionary and returns the learning result to an initial value by a predetermined amount based on the collation result. 2. The learning means according to claim 1, wherein the number of times of voice input is equal to or more than a predetermined value, and the input voice is matched with the dictionary and the number of matches between the two is equal to or less than a predetermined value. The voice recognition device according to claim 1, wherein the voice recognition device is returned to a side. 3. The speech recognition apparatus according to claim 2, wherein the learning means changes a predetermined amount for returning the learning result to the initial value side in accordance with a ratio between the number of times of voice input and the number of times of matching. 4. The speech recognition apparatus according to claim 1, wherein the learning means calculates a learning progress degree based on a learning result and displays the learning progress degree. 5. The speech recognition apparatus according to claim 4, wherein the learning means issues an alarm when the learning progress degree falls below a predetermined value.