JPH05158493A - Speech recognizing device - Google Patents

Abstract

PURPOSE:To obtain accurate recognizing result with less input frequency about sound voice information, relating to a speech recognizing device which recognizes and outputs a character string displayed by sound voice information inputted by a user. CONSTITUTION:In a voice sound recognizing device 1, the likelihood between the registered voice sound standard pattern and the sound pattern to be recognized is evaluated by obtaining local likelihood string to specify and output the character string corresponding to the voice standard pattern with highest likelihood. The device 1 is provided with a dividing part 12 which divides the character string obtained as recognizing result and divides the obtained local likelihood string in a manner corresponding to the division and a calculating part 13 which calculates the likelihood corresponding to the divided each character part by calculating a representative value of divided each local likelihood string. When the character string obtained as recognizing result is outputted, the likelihood calculated by the calculating part 13 is made to be related with corresponding character part.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention operates so as to recognize which one of pre-registered character strings corresponds to voice information input by a user and output the recognition result to the user. The present invention relates to a voice recognition device, and more particularly to a voice recognition device that can obtain an accurate recognition result with a small number of times of inputting voice information.

In recent years, a system is becoming widespread in which voice information uttered by an operator is recognized and processing such as automatic distribution of objects is executed according to the recognition result. The voice recognition device used in such a system is
While recognizing which of the registered character strings the voice information uttered by the operator corresponds to, the operator, if the recognition result is not correct, the voice information until the correct recognition result is obtained. Will be repeated. Therefore, in such a voice recognition device, it is necessary to adopt a configuration in which the recognition result desired by the operator can be obtained with a small number of voice information utterances.

[0003]

2. Description of the Related Art In a voice recognition device that recognizes which of the character strings registered in advance the voice information input by the user corresponds to and displays the recognition result to the user, Conventionally, the configuration has been adopted in which the character string of the recognition result of the first place is displayed or the character string of the recognition result of a plurality of higher ranks is displayed.

When none of the recognition results is correct, the user again speaks voice information and inputs it to the voice recognition device.

[0005]

As described above, since the conventional voice recognition device merely displays the character string which is the recognition result, the user can recognize which voice information the voice recognition device has. If there is no way of knowing whether or not a part has been erroneously recognized, and if the voice recognition device erroneously recognizes it, the user will blindly repeat the input of voice information. from now on,
There was a problem that voice information had to be input many times before the correct recognition result was obtained.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a new voice recognition apparatus which can obtain an accurate recognition result with a small number of times of inputting voice information. ..

[0007]

FIG. 1 illustrates the principle configuration of the present invention. In the figure, 1 is a voice recognition device equipped with the present invention, and 2 is an output device for outputting the recognition result of the voice recognition device 1.

The voice recognition device 1 includes a standard pattern management unit 1
0, a recognition unit 11, a division unit 12, a calculation unit 13, and an output unit 14. This standard pattern management unit 10
A voice pattern of a voice signal that may be input is managed as a voice standard pattern while having a correspondence with a character string. When a voice pattern of a voice signal to be recognized is given, the recognition unit 11 evaluates the likelihood of the voice pattern and the voice standard pattern managed by the standard pattern management unit 10 while obtaining a local likelihood sequence. The character string represented by the voice signal to be recognized is recognized.

The dividing unit 12 divides the character string recognized by the recognizing unit 11 and, corresponding to the division, the recognizing unit 11
The local likelihood sequence obtained by is divided. Dividing unit 12
In this division processing, the character string of the recognition result may be divided based on time, and the character string of the recognition result may be divided based on the phoneme or syllable.

The calculating unit 13 calculates the representative value of each local likelihood sequence divided by the dividing unit 12, and thereby the dividing unit 12
The likelihood associated with each character part divided by is calculated. The output unit 14 outputs the recognition result to the output device 2.

[0011]

In the present invention, when the voice pattern to be recognized is given to the recognition unit 11 by the user uttering voice information to be recognized, the recognition unit 11 causes the voice pattern and the standard pattern management unit 10 to operate. The likelihood with the managed voice standard pattern is evaluated while obtaining the local likelihood sequence, and the voice standard pattern with the highest likelihood or a character string corresponding to a plurality of high-rank voice standard patterns is recognized. Recognize as a character string that represents information.

When the recognition unit 11 obtains the character string of the recognition result in this way, the division unit 12 divides the character string of the recognition result based on, for example, a phoneme, and, corresponding to the division, The local likelihood sequence obtained by the recognition unit 11 is divided, and subjected to this division processing, the calculation unit 13
Calculates the likelihood associated with each character portion divided by the dividing unit 12 by calculating, for example, an average value of each local likelihood sequence divided by the dividing unit 12.

Then, when the output unit 14 receives the calculation result of the calculation unit 13 and outputs the character string of the recognition result recognized by the recognition unit 11, the calculated likelihood corresponds to each character. Output in association with the part. At this time, one or a plurality of relatively high likelihoods among the calculated likelihoods,
Either or both of the character part associated with this likelihood may be output in an output form different from the output form of the other character part.

As described above, in the voice recognition device 1 of the present invention, the likelihood of each character portion obtained by the recognition processing is displayed in association with each other, instead of simply displaying the character string as the recognition result. Therefore, the user can know which voice information part to utter when paying attention to the voice information when the voice information is re-input due to the incorrect recognition result. You don't have to type in many times.

[0015]

EXAMPLES The present invention will be described in detail below with reference to examples. FIG. 2 illustrates one embodiment of the present invention. In the figure, 1 is an embodiment of a voice recognition device according to the present invention, 2a is a display device for outputting a recognition result of the voice recognition device 1, and 3 is a word by converting word voice information uttered by a user into an electric signal. A microphone that generates an audio signal.

The voice recognition device 1 of this embodiment includes a voice input unit 15, a frequency analysis unit 16, a standard frequency pattern management unit 10a, a collation unit 11a, a phoneme division unit 12a,
The phoneme likelihood calculation unit 13a and the recognition result display control unit 14a are provided.

The voice input unit 15 includes a microphone 3
A / D conversion is performed on the converted word voice signal. The frequency analysis unit 16 frequency-analyzes the A / D-converted word voice signal by the voice input unit 15 for each appropriate period to display a time-frequency pattern (what frequency characteristic has at each time). ) Get. The standard frequency pattern management unit 10a corresponds to the standard pattern management unit 10 of FIG. 1, and associates a typical time-frequency pattern of a word voice signal that may be input with a word character string. It is managed as a standard frequency pattern while taking a relationship.

The matching unit 11a corresponds to the recognition unit 11 in FIG. 1, and uses a matching technique such as DP matching, for example, to analyze the time-frequency pattern analyzed by the frequency analysis unit 16 and the standard frequency pattern management. The reference frequency pattern managed by the unit 10a is collated, the word character string having the highest likelihood (similarity) is specified and output to the phoneme division unit 12a, and the local likelihood sequence obtained at the time of the collation is specified. The (sequence of local likelihoods on the optimum matching route) is output to the phoneme dividing unit 12a.

The phoneme dividing unit 12a corresponds to the dividing unit 12 in FIG. 1, divides the word character string obtained by the matching unit 11a into phonemes, and, in accordance with the division, the matching unit. The local likelihood sequence obtained by 11a is divided for each phoneme. The phoneme likelihood calculating unit 13a corresponds to the calculating unit 13 in FIG. 1, and calculates the average value of the local likelihood sequence for each phoneme divided by the phoneme dividing unit 12a to calculate the likelihood for each phoneme. To calculate. Recognition result display control unit 14
a corresponds to the output unit 14 in FIG. 1 and outputs the recognition result to the display device 2a.

Next, the operation processing of the embodiment thus constructed will be described in detail. When the user utters word voice information to be recognized, the voice input unit 15 A / D-converts the word voice signal converted into an electric signal by the microphone 3, and the frequency analysis unit 16 performs the A / D conversion. A time-frequency pattern is obtained by frequency-analyzing the word speech signal for each appropriate period.

When the time-frequency pattern of the voice information uttered by the user is obtained in this way, the collating unit 11a.
Compares the time-frequency pattern analyzed by the frequency analysis unit 16 with the standard frequency pattern managed by the standard frequency pattern management unit 10a, identifies the word string with the highest likelihood, and outputs it to the phoneme division unit 12a. At the same time, the local likelihood sequence obtained during the matching is output to the phoneme dividing unit 12a.

In response to the word character string output processing by the collating unit 11a, the phoneme dividing unit 12a divides the received word character string for each phoneme, and the collating unit 11a finds the corresponding word segment. The local likelihood sequence is divided for each phoneme, and the phoneme likelihood calculating unit 13a receives the local likelihood sequence dividing process by the phoneme dividing unit 12a. Is calculated to calculate the likelihood for each phoneme.

Then, the recognition result display control unit 14a displays the word character string of the recognition result recognized by the collation unit 11a on the display device 2a, and receives the calculation result of the phoneme likelihood calculation unit 13a, Control is performed so that the calculated likelihood for each phoneme is displayed in association with the corresponding phoneme of the word character string of the recognition result.

In this way, the user can select, for example, "CHI
If the collation unit 11a recognizes "SHIGA (Shiga)" when uttering the word voice information "BA (Chiba)", the phoneme segmentation unit 12a is
"/ SH /, / I /, / G /, / A /" is obtained by dividing the recognition result "SHIGA" in units of phonemes, and the phoneme likelihood calculation unit 13a uses, for example, "SH" Likelihood is "1"
It is calculated that the likelihood of “I” is “5”, the likelihood of “G” is “4”, and the likelihood of “A” is “12” with 0 ”, and these processing results are Receiving, recognition result display control unit 1
As shown in FIG. 3, 4a is a word character string “S” which is a recognition result on the display screen of the display device 2a.
The control is performed such that "HIGA" and the likelihood of each phoneme that constitutes "SHIGA" are displayed in association with each other.

According to this display control, the user can select "CHI
"CH" of "BA" is mistakenly recognized as "SH", and "CHIB"
When I know that "B" of "A" is mistakenly recognized as "G", I can know that the likelihood of "SH" is high, so when I re-enter "CHIBA" ,
It is possible to know that it is sufficient to say "CHIBA" while paying attention to "CH" that was mistaken with high likelihood.

In the display example shown in FIG. 3, the likelihood calculated by the phoneme likelihood calculating unit 13a is simply displayed in association with each phoneme of the word character string of the recognition result. However, for example, a relatively high likelihood may be displayed in a color different from other likelihoods or flicker displayed, or a phoneme associated with that likelihood may be displayed in a different color or flicker displayed. This is convenient for calling the user's attention.

As described above, in the voice recognition device 1 of the present invention, the likelihood of each character portion obtained by the recognition process is displayed in association with each other, instead of simply displaying the character string as the recognition result. It is characterized by taking.

Although the illustrated embodiment has been described, the present invention is not limited to this. For example, in the example,
Although the present invention has been described by taking the recognition processing of word voice information as an example, the present invention is not limited to this, and can be applied as it is to the voice recognition processing of phrases and the like.

Further, although the embodiment has been disclosed by dividing the character string of the recognition result based on the phoneme, the present invention is not limited to this, and the recognition result of the recognition result is based on the syllable or the time. It is also possible to adopt a configuration in which the character string is divided. For example, when divided based on time, information that the user should pay attention to the first half of the voice information to be uttered or the second half of the voice information should be displayed.

[0030]

As described above, according to the present invention,
In the voice recognition device, when displaying the character string of the recognition result, since the likelihood of each character portion obtained by the recognition processing is associated and displayed, the user is Since it is an error, you can re-enter the voice information and know which voice information part should be uttered, so that you do not have to input the voice information many times. It will be.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is an example of the present invention.

FIG. 3 is an explanatory diagram of a display example of a recognition result.

[Explanation of symbols]

 1 voice recognition device 2 output device 10 standard pattern management unit 11 recognition unit 12 division unit 13 calculation unit 14 output unit

Claims (4)

[Claims] 1. A likelihood standard between a pre-registered voice standard pattern and a voice pattern to be recognized is evaluated while obtaining a local likelihood sequence, and a voice standard pattern having the highest likelihood or a plurality of higher-ranked voice standard patterns are evaluated. In a voice recognition device that specifies and outputs a character string corresponding to a voice standard pattern, the character string of the recognition result is divided, and the division unit that divides the obtained local likelihood sequence corresponding to the division ( 12) and by calculating a representative value of each local likelihood sequence divided by the division unit (12), the likelihood associated with each character portion divided by the division unit (12) is calculated. The calculation unit (13) is provided, and when outputting the character string of the recognition result, the calculation unit (1
A speech recognition device characterized by processing such that the likelihood calculated in 3) is output in association with a corresponding character portion. 2. The voice recognition device according to claim 1, wherein the dividing unit (12) performs processing so as to divide the character string of the recognition result on the basis of time. 3. The voice recognition device according to claim 1, wherein the dividing unit (12) performs processing so as to divide the character string of the recognition result based on a phoneme or a syllable. .. 4. The speech recognition apparatus according to claim 1, 2 or 3, wherein when outputting a character string of a recognition result, the likelihood calculated by the calculation unit (13) is relatively high. Processing such that one or a plurality of likelihoods and / or one or both of the character parts associated with the likelihoods are output in an output form different from the output form for the other character parts. , Characteristic voice recognition device.