JP2555029B2 - Voice recognition device - Google Patents

Description

The present invention relates to a voice recognition device, and more particularly to a voice recognition device suitable for improving a recognition rate.

[Conventional technology]

The conventional specific speaker recognition device registers a characteristic pattern of a voice to be recognized as a standard voice pattern in advance, compares the characteristic pattern of the voice input during the recognition operation with the standard voice pattern described above, and determines the most similar standard. It is based on the method of selecting a voice pattern as a result of voice recognition.

However, the voice uttered by a human changes due to various reasons such as changes in physiological organs, particularly vocal cords and throat over time, health condition, and mental condition. That is, its characteristic pattern is variable.

Therefore, even if the speaker intends to say the same word as the registered voice, the similarity as a result of the pattern comparison varies. When the width of this variation is large, there are cases where the input feature pattern has a greater degree of similarity with other standard patterns of the registered voice. As a result, a voice different from the input voice is selected (erroneous recognition), which is a major cause of reduction in the recognition rate.

As a method for preventing this, for example, JP-A-60-218698
In the standard voice pattern registration process, the input pattern to be registered is compared with the other standard patterns registered up to that point, and if a similar standard pattern exists, Has given a warning to avoid registration of similar standard patterns and improve the recognition rate.

[Problems to be solved by the invention]

In the above-mentioned conventional technology, when a voice similar to the voice that has already been registered is about to be newly registered, a similar registration number is simply displayed and a warning is given. Therefore, the user must manage the correspondence between the registration number and the registered voice. This has a problem that management is troublesome when the number of registered items is large.

An object of the present invention is to efficiently and effectively perform the work of registering a standard voice pattern of a speaker and aim to improve the recognition rate.

[Means for solving problems]

In order to achieve the above object, the present invention compares a speaker's input voice characteristic pattern with a standard voice characteristic pattern of the speaker registered in advance, thereby recognizing the input voice pattern. The person voice recognition device has means for recording and reproducing the standard voice at the time of registering the standard voice feature pattern, and at the time of registering the standard voice feature pattern, the input voice feature pattern of the speaker and the voice pattern previously registered. If a plurality of standard voices having a certain degree of similarity are compared with the standard voice feature pattern, the standard voice at the time of registration of the previously registered standard voice pattern is similar when reproduced by the recording / reproducing means. It is characterized in that reproduction is performed in descending order of degree.

[Action]

At the time of voice registration, a characteristic pattern is extracted from a voice input to be registered, and is compared with a standard voice pattern of another voice previously registered. As a result, when there is a registered voice having a certain degree of similarity or more, the registration of the input voice is stopped and the similar registered voice is uttered. This allows the user to know which registered voice is similar to the registered voice, which is useful for re-inputting a new registered voice. Therefore, the registration work can be efficiently and effectively performed without considering the correspondence between the registration number and the registration voice as in the conventional case, and the recognition rate can be improved.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS.

FIG. 1 is a block diagram showing the configuration of the present embodiment, where 1 is a micro processing unit (hereinafter
2) Microphone, 3 A / D converter, 4 speaker, 5 D / A converter, 6 random access memory (hereinafter referred to as RAM), 7 read-only
Memory (hereinafter referred to as ROM), 8 is a voice feature pattern extraction unit, 9 is a pattern comparison unit, and 10 is a standard pattern storage RA.
M and 13 are registration / recognition switching devices. The RAMs 6 and 10 are backed up by a battery so that the contents can be retained even when the power is off.

FIG. 2 is a flowchart explaining the operation of the MPU 1 of FIG. FIG. 3 is a flowchart explaining the operation at the time of voice recognition in FIG.

The operation will be described below according to this flowchart. After starting, the MPU 1 first determines whether the registration is recognized (step A). This is done by checking the state of the registration / recognition switching device 13 in FIG. At the time of registration, a voicing instruction sound that prompts the user to utter a registered voice is generated (step B). The voicing instruction sound is generated from the speaker 4 after the MPU 1 reads the waveform data of the sound recorded in advance in the ROM 7 and gives it to the D / A converter 5 to convert it into an analog signal. Next, the MPU 1 enters a state of waiting for input of registered voice (step C). When the user speaks, the microphone 2
Is converted into an analog electric signal and further converted into digital data by the A / D converter 3. This data is read by the MPU 1 (step D) and its size is monitored (step E). Then, for example, if the volume is a certain level or more, it is determined that a sound is input, and the subsequent data is stored in the RAM 7 and at the same time given to the characteristic pattern extraction unit 8. At the end of voice, for example, if a silent section continues for 300 ms or more,
It is determined that the voice input is completed, and the voice data acquisition is ended. Next, when the output from the characteristic parameter extraction unit 8 is received (step F) and there is no previously registered voice,
This characteristic pattern is registered in the RAM 10 as a standard pattern. At the same time, a record in which the storage location of the waveform data of the voice stored in the RAM 6 and the storage location of the standard pattern are recorded together is created on the RAM 6 (step J).
If there is a previously registered voice, the feature patterns of all the voices registered as standard patterns up to that time and the output from the above-mentioned feature parameter extraction unit 8 are sequentially compared by the comparator 9. (Step H). If there is no registered voice having a degree of similarity equal to or higher than a certain value, the process directly proceeds to step J and proceeds to the registration operation described above. However, if there is a registered voice having a degree of similarity equal to or higher than a certain value, a sound is generated by the same means as in step B (step K), and then the registered voice is reproduced (step L). When there are a plurality of similar registration voices, all of them may be reproduced in order of similarity or randomly. Alternatively, one having the highest degree of similarity may be reproduced. In this method, the waveform data of the voice stored in the RAM 6 is read at a constant cycle and transferred to the D / A converter 5. After being converted into an analog electric signal by the D / A converter 5, it is reproduced as a voice by the speaker 4. After this, the procedure returns to the step B and shifts to the re-registration state.

Next, the operation at the time of recognition will be described. After branching to the recognition operation in step A of FIG.
Enter). Here, first, a voice input instruction sound is generated (step N), and a voice input waiting state is entered (step O). When the voice is input, the voice data is sent to the characteristic pattern extraction unit 8 to extract the characteristic pattern (step R). This characteristic pattern is compared with the standard pattern of the registered voice stored in the RAM 10 in the successive comparator 9,
The degree of similarity is calculated (step S). Then, the registered voice with the highest similarity is determined as the recognition result from the obtained similarities of all the registered voices (step T). Further, as a determination result, the registered voice is reproduced (step U), and the result processing routine is entered (step V). Here, for example, when the recognition device is connected to an external device via an interface, the recognition result is sent to the device.

Further, with the configuration of the present embodiment, various functions can be created by changing the program of MPU1.
In FIG. 4, the utterance of the input voice (step L ′) is performed after the utterance of the similar voice (step L) in FIG. 2. That is, at the time of registration, if a voice similar to the voice previously registered is input to register, the already registered similar voice and the input voice are continuously uttered and the re-input is urged without performing the registration. In this case, the order of step L and step L'may be exchanged.

Further, in the present embodiment, the method of newly registering with the registered voice left was described, but on the contrary, the similar registered voice is deleted, and instead, the data and the characteristic pattern of the newly input voice are replaced. The function of registering is also possible. A method of dividing the registered voice and the standard voice pattern into several groups and checking for similar voices only within the same group is also conceivable.

As a method of storing registered voice, PCM (Pulse Code Modulation), ADPCM (Adaptive Differential Pulse Coat Modulation), etc. can be considered, but let's reduce the memory cost by performing advanced voice compression. Then, the quality of the reproduced sound tends to deteriorate. Therefore, an appropriate method may be used in consideration of factors such as the intended price of the device. Various choices can be made for the voice instruction sound, such as means, method, and tone color, but a method using voice synthesis is also possible and a smooth operation guide can be realized. Since all the means for realizing it are included in the configuration described in the embodiment, it can be easily realized.

Further, the storage medium itself is not limited to a semiconductor memory or the like having a fast access time, but a storage medium using a magnetic recording system such as a floppy disk device is also conceivable.

According to this embodiment, when a voice having a characteristic pattern similar to the previously registered standard pattern is registered at the time of registering the voice standard pattern, the registration is prohibited. Further, since the similar already-registered voice is reproduced to prompt re-registration, confirmation of the similar voice can be easily performed, and it can be used as a reference for re-registration and smooth registration work can be performed.

〔The invention's effect〕

According to the present invention, in the specific-speaker voice recognition device, when an attempt is made to register a voice similar to the previously registered voice during the work of registering a word to be recognized, a similar registered voice is uttered and re-issued to the user. You can prompt for input.
In particular, by uttering a similar registered voice,
The user can effectively confirm which of the registered voices is similar to the registered voice, which will be a reference for the next registration. Therefore, the registration work can be performed efficiently, and the registration of similar voices can be prevented in advance, which has an effect of improving the recognition rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention. FIG. 2 is a flow chart showing an operation flow of the MPU 1 in FIG. 1, and FIG. 3 is a flow chart showing a recognition operation in FIG. Flow diagram,
FIG. 4 is a flow chart showing an embodiment in which a function of reproducing an input voice is added to the embodiment shown in FIG. 1 ... MPU, 2 ... Microphone, 3 ... A / D converter, 4 ... Speaker, 5 ... D / A converter, 6 ... ROM, 7 ... RAM, 8 ... Audio feature pattern extraction unit, 9 ... Pattern comparator, 10 ... RAM for storing standard patterns.

Continuation of the front page (56) Reference JP 62-148998 (JP, A) JP 61-3241 (JP, A) JP 60-59395 (JP, A) JP 60-218698 (JP , A) JP 57-141700 (JP, A) JP 57-129497 (JP, A) JP 60-201396 (JP, A)

Claims (2)

(57) [Claims] 1. A specific speaker voice recognition device for recognizing an input voice pattern by comparing an input voice feature pattern of a speaker with a standard voice feature pattern of the speaker registered in advance, It has means for recording and reproducing the standard voice at the time of registration of the standard voice feature pattern, and at the time of registering the standard voice feature pattern, compares the input voice feature pattern of the speaker with the previously registered standard voice feature pattern. However, when there are a plurality of standard voices having a certain degree of similarity or more, the standard voices at the time of registration of the previously registered standard voice pattern are reproduced in the descending order of similarity when being reproduced by the recording / reproducing means. A voice recognition device. 2. The voice recognition device according to claim 1, wherein the input voice at the time of registration and the standard at the time of registration of the previously registered standard voice pattern, when the voice recognizer has a similarity of a certain level or more. A voice recognition device characterized by continuously reproducing voice.