JPH11352988A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice recognition device capable of efficiently finding out an erroneously recognized character. SOLUTION: This voice recognition device is provided with a voice recognition part 2 for recognizing the voice data inputted from a voice data input part 1 in reference to a dictionary 3 and converting it to a test data; an erroneously recognized character estimation part 4 for determining the difference in likelihood between a recognition determined character and the character which is the second candidate in recognition determination by use of the likelihood of each character obtained from the voice recognition processing part 2 and estimating the character having a difference of a prescribed threshold or less as a character having a high erroneous recognition probability; a parameter input part 5 for setting the threshold; a display processing part 6 for performing the processing of changing and displaying the color of form of the erroneously recognized character estimated by the erroneously recognized character estimation part 4 to characters other than it; and a display part 8 for displaying the processing result by the display processing part 6.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device, and more particularly, to a voice recognition device that recognizes voice data, converts the voice data into text data, and displays the text data.

[0002]

2. Description of the Related Art A so-called voice word processor, or a dictation system that automatically creates a document based on voice data when inputting voice data by dictation and displays the document on a screen or the like, has conventionally been realized. This is one of the goals in speech recognition system development, and research and development are currently being actively pursued.

With the recent progress of speech recognition technology, a microphone is connected to a personal computer,
A device has been developed in which voice input using this microphone is documented on the personal computer and displayed on a screen, and is generally commercially available.

On the other hand, conventionally, when a document is created, the content of the document to be created is once dictated and recorded on a recording device such as a tape recorder, and a secretary or a typist or the like later reproduces the dictated content, and a typewriter. The use of a recording device such as a tape recorder or the like has become common as one of effective forms of using a recording device such as a tape recorder.

[0005] In the above-mentioned dictation using a recording device, there has been a strong demand for a technique for automatically converting recorded contents into a document.

Therefore, once a voice is recorded on a recording device,
The recorded content is continuously reproduced and input to a voice recognition device. The voice recognition device recognizes the continuous voice and converts it into a document. After the reproduction is completed, a document as a result of the recognition is displayed on a screen. Something to do is suggested.

[0007] In such automatic speech recognition by the speech recognition device, it is difficult to accurately convert all of the input speech into the original document, and the erroneously recognized characters are included in the converted text. Therefore, the user needs to confirm the converted document again and perform a correction operation.

Japanese Patent Publication No. 63-33174 discloses an example of a technique which makes it possible to easily perform this correction work. Means for detecting the end of a phrase or sentence, which is an input unit of the input voice by the reproducing operation of the recording device, and temporarily stopping the recording device when the end of the phrase or sentence is detected by the detecting means. Means for outputting a signal, and suspending the sound reproduction operation of the recording device in response to the pause signal,
Also, there is described a voice input type Japanese document processing device including a recording device control section for releasing the pause of the voice reproduction operation in response to the operation of a restart key.

[0009]

However, in the conventional speech recognition apparatus as described above, when checking and correcting the converted document, the user must check each character one by one. And labor was required with a heavy burden.
For this reason, when the recorded content is large and the document is long, it takes a lot of time to find all the misrecognized characters, and the work efficiency is not good.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a speech recognition device that can efficiently search for misrecognized characters.

[0011]

In order to achieve the above object, a voice recognition device according to a first aspect of the present invention includes a voice recognition processing unit that recognizes input voice data and converts the voice data into text data. A misrecognized character estimator for estimating a character with a high probability of misrecognition using the likelihood of each character generated in the course of speech recognition processing by the recognition processor, and a misrecognized character estimated by the misrecognized character estimator A display processing unit that processes the text data so that the text data can be quickly and visually identified with respect to other characters, and a display unit that displays a processing result by the display processing unit. It is.

Further, the speech recognition apparatus according to the second invention is
In the speech recognition apparatus according to the first aspect, the misrecognized character estimation unit compares the likelihood of the recognized character with a predetermined threshold value, and determines a character having a likelihood equal to or less than the threshold value with a high false recognition probability. It is estimated as a character.

Further, the speech recognition apparatus according to a third aspect of the present invention is the speech recognition apparatus according to the first aspect, wherein the erroneously recognized character estimating section recognizes the likelihood of the recognized character and determines the next candidate when the recognition is determined. Then, a difference between the likelihood of the character and the character whose difference is equal to or less than a predetermined threshold value is estimated as a character having a high probability of erroneous recognition.

Therefore, the speech recognition apparatus according to the first invention recognizes the speech data inputted by the speech recognition processing section and converts the speech data into text data, and the erroneously recognized character estimation section performs the speech recognition processing by the speech recognition processing section. Using the likelihood of each character generated in the process of estimating a character with a high probability of misrecognition, the display processing unit visually recognizes the misrecognized character estimated by the misrecognized character estimation unit with respect to other characters. Processing is performed to display the text data so that the text data can be identified as quickly as possible, and the display unit displays the processing result by the display processing unit.

[0015] Further, the speech recognition apparatus according to the second aspect of the present invention comprises:
The misrecognized character estimation unit compares the likelihood of the character determined to be recognized with a predetermined threshold, and estimates a character having a likelihood equal to or less than the threshold as a character having a high misrecognition probability.

Further, in the speech recognition apparatus according to the third invention, the erroneously recognized character estimating section may calculate a difference between the likelihood of the character determined to be recognized and the likelihood of the character which was the next candidate when the recognition was determined. Is calculated, and a character whose difference is equal to or smaller than a predetermined threshold is estimated as a character having a high probability of erroneous recognition.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. 1 to 4 show an embodiment of the present invention. FIG. 1 is a block diagram showing a configuration of a speech recognition device, FIG. 2 is a flowchart showing processing in a speech recognition processing unit, and FIG. FIG. 4 is a flowchart showing processing in the character estimating unit, and FIG. 4 is a flowchart showing processing in the display processing unit.

The speech recognition processing apparatus is configured as a personal computer that performs processing by a speech recognition processing program, for example, and inputs speech or speech data to be recognized and documented and converts the speech data into digital data. An input unit 1, a voice recognition processing unit 2 that receives an output of the voice data input unit 1 and performs voice recognition processing to convert the data to text data, and a language model referred to when the voice recognition processing unit 2 performs voice recognition And a misrecognized character estimating unit 4 for estimating a character having a high probability of misrecognition using the likelihood of each character generated in the process of speech recognition processing by the speech recognition processing unit 2. A parameter input unit 5 for setting a threshold for estimation by the erroneously recognized character estimating unit 4 and a working memory of the erroneously recognized character estimating unit 4. A memory 9, the voice recognition processing section 2
In response to the processing result of the erroneously recognized character estimating unit 4, processing is performed to display the text data so that the character estimated as the erroneously recognized character can be quickly and visually identified from other characters. A display processing unit 6;
A memory 10 serving as a work memory for the display processing unit 6
And a display unit 8 for displaying text data that has been subjected to speech recognition in response to the output of the display unit 8. The display unit 8 is used to correct a document while viewing the display on the display unit 8. The correction input is output to the display processing unit 6. And a display operation input unit 7 to be operated.

The audio data input unit 1 includes, for example, a microphone for converting an audio signal into an electric signal and an A / D for discretizing an input signal of the microphone and converting the input signal into digital data.
It may be configured to include a converter, and as another example, it may be configured to include an input device or the like that allows a recording medium on which audio data is recorded to be freely attached.

The dictionary 3 stores a standard pattern of model speech feature amounts based on a learning speech database and a statistical language model (with reading data) on the order of occurrence of characters based on a learning text database. Since it has a learning function, it is stored in a readable and writable recording medium such as a hard disk.

The speech recognition processing section 2, the misrecognized character estimation section 4, and the display processing section 6 are constituted by, for example, a CPU or the like, and sequentially execute predetermined processing programs by the CPU.

The parameter input section 5 is means for allowing a user to input and set a threshold value used when the erroneously recognized character estimating section 4 determines the magnitude of the difference between the total likelihoods described later, For example, it is configured by a mouse, a keyboard, and the like.

The display operation input unit 7 is an input means used for correcting a document while looking at the display unit 8, and is composed of, for example, a mouse and a keyboard. The input unit 5 may also be used.

The display section 8 is composed of, for example, a CRT display device or an LCD display device.

The memory 9 and the memory 10 are composed of a RAM or the like. In FIG. 1, they are shown as being separate from each other, but they may be the same memory.

Next, the operation of such an embodiment will be described.

The voice data is converted into digital data by performing predetermined sampling in the voice data input unit 1 and transmitted to the voice recognition processing unit 2.

The speech recognition process performed by the speech recognition processing section 2 will be described with reference to FIG.

When receiving the voice data from the voice data input unit 1 (step S1), the voice recognition processing unit 2 first analyzes the data and extracts voice feature amounts such as linear prediction coefficients and cepstrum coefficients. (Step S2).

Next, reference is made to the standard pattern of the feature amount of the model voice stored in the dictionary 3 (step S).
3) Select a plurality of model voices having a high similar likelihood with the input voice (step S4).

Further, with respect to the selected model voices, a statistical language model relating to the order of occurrence of characters stored in the dictionary 3 is referred to (step S5), and the probability of occurrence of each character candidate is obtained. (Step S6).

Finally, the product of the similar likelihood and the occurrence probability is set as the overall likelihood of the character candidate (step S
7) A character candidate having the largest value is determined as a result of speech recognition (step S8).

Thereafter, the speech recognition result is converted into text data (step S9) and transmitted to the display processing unit 6 (step S10).

When performing voice recognition, the above-described step S
The data of the total likelihood obtained in step 7 is transmitted to the erroneously recognized character estimation unit 4 (step S11).

Then, it is determined whether or not the voice data has been completed (step S12). If the voice data has not been completed, the flow returns to step S1 to repeat the above-described operation until the input of the voice data is completed. When the voice data is no longer input, the voice recognition process is terminated.

Next, the processing in the erroneously recognized character estimating section 4 will be described with reference to FIG.

The erroneously recognized character estimation unit 4 receives the data of the total likelihood from the speech recognition processing unit 2 (step S2).
1) The data is written into the memory 9 (step S22), and the total likelihood of the finally determined character is calculated as follows:
The difference from the overall likelihood of the character that was the next candidate when making the character recognition decision is calculated (step S23).

Then, it is determined whether or not this difference is smaller than a predetermined threshold (step S24).
It is estimated that the character has a high probability of erroneous recognition, and the position data of the character is transmitted to the display processing unit 6 (step S).
25).

If step S25 is completed or if the difference is equal to or larger than the predetermined threshold value in step S24, the processing in the erroneously recognized character estimating section 4 is completed.

Subsequently, referring to FIG.
Will be described.

The display processing unit 6 receives the text data as the recognition result of the input voice from the voice recognition processing unit 2 (step S31).
When the position data and the like of the misrecognized character estimated from are received (step S32), these data are written into the memory 10 (step S33).

Then, the display color of the estimated misrecognized character (or the background color of the estimated misrecognized character) is changed to a color different from the display color of the other characters (or the background color of the other characters). Then, processing is performed so that these display colors are displayed on the display unit 8 at a subsequent stage (step S3).
4), the processing in the display processing unit 6 ends.

As a result of performing such processing, a character displayed in a different color from other characters appears in, for example, several places in the displayed document on the display unit 8, and this is a character that is assumed to be a misrecognized character. Becomes

Thus, the user can easily confirm a character presumed to be a misrecognized character by a different display color, and determines whether or not the character is actually a mistake, and determines that the character is a mistake. In this case, the correction is made by re-inputting a correct character from the display operation input unit 7.

In addition, most of the documents displayed on the display unit 8 are characters having different colors which are presumed to be erroneously recognized characters, or characters having different colors which are presumed to be erroneously recognized characters do not exist at all. In such a case, there is a possibility that the setting of the threshold value is not appropriate. Therefore, the user resets the threshold value using the parameter input unit 5.

By setting an appropriate threshold value by performing trial and error several times in this way, it is possible to estimate a misrecognized character with a high probability and to quickly find the position of the character visually. The document can be corrected well.

In the erroneously recognized character estimating unit 4 described above, a threshold value is provided for the difference between the total likelihood of a character determined to be recognized and the character of the next candidate character. The present invention is not limited to this. For example, a threshold value may be provided for the overall likelihood of a character determined to be recognized, and a character less than the threshold value may be estimated as a misrecognized character, or both of them may be used. May be performed for the estimation.

In the above description, the threshold value used in the erroneously recognized character estimation unit 4 is set by the user himself through the parameter input unit 5. May be. That is, an initial value may be prepared as the threshold value, and the initial value may be used when the setting is not performed by the user. Furthermore, the speech recognition device itself determines the degree of occurrence of the character that is estimated to be erroneously recognized in the entire document, automatically changes the threshold value by appropriate steps, and automatically finds the optimum threshold value. The configuration may be such that

In the above description, the color is changed so that the estimated misrecognized character can be quickly and visually distinguished from other characters. However, the present invention is not limited to this. Not a thing, but a shape change,
It goes without saying that, for example, means for changing the typeface, changing the character size, and underlining the corresponding portion may be used.

Further, the total likelihood of the estimated misrecognized character is subdivided into a plurality of stages, and the color scheme, shading, and the like are changed according to the stage to specify the overall likelihood. However, it may be possible to distinguish between those with a higher probability of misrecognition and those with a relatively low probability of misrecognition.

According to the speech recognition apparatus of such an embodiment, a character having a high probability of misrecognition is estimated from the likelihood generated during speech recognition, and the estimated misrecognized character is displayed differently from other characters. Since the method is displayed on the screen by the method, when checking the document, the user can quickly find and confirm the erroneously recognized character visually, and can efficiently correct the document. .

It should be noted that the present invention is not limited to the above-described embodiments, and it is needless to say that various modifications and applications can be made without departing from the gist of the invention.

[Appendix] According to the above-described embodiment of the present invention as described in detail above, the following configuration can be obtained.

(1) Speech recognition processing unit for recognizing input speech data and converting it to text data, and erroneous recognition using the likelihood of each character generated in the course of speech recognition processing by the speech recognition processing unit A misrecognized character estimating unit that estimates a character having a high probability of, and the misrecognized character estimated by the misrecognized character estimating unit, so that the other characters can be quickly and visually identified. A speech recognition device comprising: a display processing unit that processes text data to be displayed; and a display unit that displays a processing result of the display processing unit.

(2) The misrecognized character estimating section compares the likelihood of a character determined to be recognized with a predetermined threshold value, and estimates a character having a likelihood equal to or smaller than the threshold value as a character having a high probability of misrecognition. The speech recognition device according to supplementary note (1), wherein the speech recognition device is a speech recognition device.

(3) The erroneously recognized character estimating section obtains a difference between the likelihood of the character determined to be recognized and the likelihood of the character which is the next candidate when the recognition is determined, and determines this difference as a predetermined value. The speech recognition apparatus according to (1), wherein a character having a threshold value or less is estimated as a character having a high misrecognition probability.

(4) The speech recognition apparatus according to the additional note (2) or (3), further comprising a parameter input section for setting the threshold value.

(5) The display processing section is characterized in that the estimated misrecognized character can be visually and quickly identified by changing the color of the other characters. (1).

(6) The display processing section further changes the color according to the probability that the estimated misrecognized character is misrecognized. Voice recognition device.

(7) The display processing section is characterized in that the estimated misrecognized character can be visually and quickly identified by changing the shape of other estimated characters. (1).

Therefore, according to the invention described in the appendix (1), the erroneously recognized character is displayed so as to be visually identifiable quickly with respect to other characters, so that the erroneously recognized character is efficiently displayed. You can find characters.

According to the invention described in the supplementary note (2), a character whose likelihood is equal to or less than a predetermined threshold value among the characters determined to be recognized is estimated as a character having a high misrecognition probability. The same effect as the invention described in (1) can be obtained.

According to the invention described in the appendix (3), a character whose difference between the likelihood of the character determined to be recognized and the likelihood of the character which is the next candidate at the time of recognition determined is equal to or smaller than a predetermined threshold value. Is estimated as a character having a high misrecognition probability, the same effect as that of the invention described in Appendix (1) can be obtained.

According to the invention described in Supplementary Note (4), the same effects as those of the invention described in Supplementary Note (2) or (3) can be obtained, and since the parameter input unit is provided, the user sets the threshold value. It becomes possible to input and an optimal threshold can be selected.

According to the invention described in Supplementary Note (5), the same effect as the invention described in Supplementary Note (1) can be obtained, and the color of the estimated misrecognized character can be changed for other characters. This facilitates visual identification.

According to the invention described in the supplementary note (6), the same effect as the invention described in the supplementary note (5) can be obtained, and the color is changed according to the probability that the estimated erroneously recognized character is erroneously recognized. Therefore, it is possible to grasp the certainty that the recognition is erroneous.

According to the invention described in Supplementary Note (7), the same effect as that of the invention described in Supplementary Note (1) is obtained, and the shape of the erroneously recognized character is changed with respect to other characters. This facilitates visual identification.

[0068]

As described above, according to the speech recognition apparatus of the first aspect of the present invention, an erroneously recognized character is displayed so that it can be visually identified quickly from other characters. In addition, it is possible to efficiently search for misrecognized characters.

Further, according to the speech recognition apparatus of the present invention, a character whose likelihood is equal to or less than a predetermined threshold value among the characters determined to be recognized is estimated as a character having a high erroneous recognition probability. The same effects as those of the first aspect can be obtained.

Further, according to the speech recognition apparatus of the present invention, the difference between the likelihood of the character determined and recognized and the likelihood of the character that was the next candidate when the recognition was determined is determined by the predetermined threshold value. By estimating the following characters as characters having a high misrecognition probability, the same effects as those of the first aspect can be obtained.

[Brief description of the drawings]

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

FIG. 2 is a flowchart showing processing in a voice recognition processing unit of the voice recognition device of the embodiment.

FIG. 3 is a flowchart showing a process in a misrecognized character estimation unit of the speech recognition device of the embodiment.

FIG. 4 is a flowchart showing processing in a display processing unit of the voice recognition device of the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Speech data input part 2 ... Speech recognition processing part 3 ... Dictionary 4 ... Error recognition character estimation part 5 ... Parameter input part 6 ... Display processing part 7 ... Display operation input part 8 ... Display part 9, 10 ... Memory

Claims (3)

[Claims] 1. A speech recognition processing unit for speech recognition of input speech data to convert the speech data into text data, and an erroneous recognition using the likelihood of each character generated in a speech recognition process by the speech recognition processing unit. A misrecognized character estimator for estimating a character having a high probability, and a misrecognized character estimated by the misrecognized character estimator,
A display processing unit that processes the text data to be displayed so that other characters can be quickly and visually identified; and a display unit that displays the processing result of the display processing unit. A speech recognition device characterized by the above-mentioned. 2. The erroneously recognized character estimating unit compares a likelihood of a character determined to be recognized with a predetermined threshold, and estimates a character having a likelihood equal to or smaller than the threshold as a character having a high erroneous recognition probability. The speech recognition device according to claim 1, wherein 3. The erroneously recognized character estimating unit obtains a difference between the likelihood of the character determined to be recognized and the likelihood of the character that is the next candidate when the recognition is determined, and determines the difference as a predetermined threshold. The speech recognition apparatus according to claim 1, wherein the following characters are estimated as characters having a high misrecognition probability.