KR20080066243A - Voice recognition device and method of voice recognition using the same - Google Patents

Abstract

An apparatus and a method for recognizing voice are provided to perform voice recognition using various parameters, thereby improving the efficiency of the voice recognition. A voice-receiving unit(110) receives a voice signal, and converts the voice signal into voice recognition data having main parameter and sub parameter for identifying the voice signal. A pattern-storing unit(130) stores predetermined voice model data patterns, and a feature-extracting unit(120) extracts a recognition feature part of the voice recognition data according to the main parameter and the sub parameter. If the recognition feature part is identical to one of the voice model data patterns, a controller(150) outputs a command code corresponding to the voice model pattern.

Description

VOICE RECOGNITION DEVICE AND METHOD OF VOICE RECOGNITION USING THE SAME}

1 is a flowchart illustrating a general speech recognition method.

2 is a diagram illustrating an example of converting an analog voice signal into digital voice recognition data.

3 is a block diagram illustrating a speech recognition apparatus according to an embodiment of the present invention.

4 is a diagram for describing a speech recognition data structure having directionality.

5 is a view for comparing and explaining a general speech recognition data structure and a speech recognition data structure according to the present invention.

6 is a flowchart illustrating a speech recognition method according to an embodiment of the present invention.

* Description of the symbols for the main parts of the drawings *

100: speech recognition device 110: voice receiver

120: feature extraction unit 130: pattern storage unit

140: command code storage unit 150: control unit

The present invention relates to a speech recognition apparatus and a speech recognition method using the same, and more particularly, to a speech recognition apparatus having an improved speech recognition rate and a speech recognition method using the same.

Recently, in order to improve the convenience of operation of an electronic device, a voice recognition technology for driving a predetermined function of an electronic device by directly inputting a command through a user's voice without a user directly touching a command key formed on the electronic device has been performed. It is developing.

The speech recognition technology is divided into a speaker-dependent system that recognizes only a specific user and a speaker-independent system that recognizes a specific user.

Speaker dependent speech recognition stores and registers the speaker's voice before use and recognizes the voice by comparing the pattern of the input voice with the stored voice. On the other hand, speaker independent speech recognition is for recognizing an unspecified majority of speakers' voices, and there is no need for a user to register a voice before operating a system like speaker dependent speech recognition.

That is, the voice of unspecified majority speakers is collected to store statistical voice model data, and speech recognition is performed using the voice data. Therefore, the characteristic characteristics of each speaker disappear, and the characteristics common to each speaker are highlighted. Speaker-dependent speech recognition has a higher recognition rate and easier implementation of technology than speaker-independent speech recognition.

A method of performing speech recognition using the speech recognition technology will now be described in detail.

1 is a flowchart illustrating a general voice recognition method, and FIG. 2 is a diagram illustrating an example of converting an analog voice signal into digital voice recognition data.

1 and 2, the general speech recognition method includes receiving a speech signal (S10), converting the speech signal into speech recognition data having only a magnitude (S20), a noise processing step (S30), and speech recognition. Extracting a recognition feature of the data (S40), comparing the speech model data pattern and the recognition feature (S50), determining whether a matching pattern exists (S60), and outputting a recognition result (S70). It includes.

Specifically, when the user outputs a predetermined command as a voice signal in step S10, an acoustic sensor such as a microphone receives a voice signal output by the user.

In step S20, the acoustic sensor converts the digital voice signal (hereinafter, voice recognition data) as shown in FIG. 2 using the received analog voice signal.

The speech component included in the speech recognition data converted in step S30 and separated from the speech recognition data for the noise component input from the surrounding environment and the speech signal input from the actual user, to remove the noise component.

In operation S40, a recognition feature capable of easily distinguishing the noise-recognized speech recognition data, for example, a speech signal is classified into phoneme units, and a recognition feature such as a collection of classified phonemes is extracted.

The recognition feature extracted in step S50 is compared with the speech model data pattern. Here, the voice model data pattern means a pattern registered by learning in both the speaker dependent system and the speaker independent system described above, or by modeling common features of voice signals of unspecified majority speakers. It should be pre-stored in a database (hereinafter referred to as a voice DB).

In operation S60, it is determined whether a comparison result of the speech model data pattern and the extracted recognition features is present, and whether there is a pattern that matches the recognition features, and in operation S70, the recognition result is output. Here, if the recognition result is a voice model data pattern that matches the recognition feature, a command code previously matched with the voice model data pattern is output so that the electronic device corresponds to the command code, that is, corresponding to the command output by the user. Perform a function.

According to this general speech recognition method, a speech signal is converted into speech recognition data having only a magnitude, a recognition feature is extracted on the basis of the speech signal, and the speech is compared with the previously stored speech model data pattern and the recognition feature in consideration of the magnitude of the speech signal. There is a problem in that the speech recognition rate is lowered as the recognition is performed.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to provide a speech recognition apparatus having improved speech recognition rate by performing speech recognition using various parameters.

Another object of the present invention is to provide a voice recognition method using the voice recognition device.

In order to achieve the above object of the present invention, a voice recognition apparatus according to an embodiment of the present invention receives a voice signal and converts it to voice recognition data having a main parameter and a sub parameter for identifying the voice signal A pattern storage unit for storing a predefined speech model data pattern, a feature extractor extracting a recognition feature of the speech recognition data according to the main and sub-parameters, and comparing the recognition feature with the speech model data pattern And if there is a speech model data pattern that matches a recognition feature, then outputting a command code corresponding to the speech model data pattern.

Here, the main parameter may be magnitude information of the voice signal, and the sub parameter may be direction information on an output direction of the voice signal.

The voice receiver may include a plurality of acoustic sensors that identify the direction information in a 3D space region and include the direction information in the voice recognition data.

Preferably, the voice recognition apparatus according to an embodiment of the present invention may further include a command code storage unit for storing the command code.

The pattern storage unit may store predefined voice model data patterns for the voice signal according to the main parameter and the sub parameter.

In order to achieve another object of the present invention, the voice recognition method according to an embodiment of the present invention comprises the steps of receiving a voice signal and converting it into voice recognition data having a main parameter and a sub-parameter for identifying the voice signal, Extracting a recognition feature of the converted speech recognition data, comparing a predefined speech model data pattern with the recognition feature, and if it is determined that the speech model data pattern coincides with the recognition feature, the speech Outputting a command code corresponding to the model data pattern.

Here, the main parameter may be magnitude information of the voice signal, and the sub parameter may be direction information on an output direction of the voice signal.

According to the display device according to the present invention, the speech recognition rate can be improved by performing speech recognition by comparing various samples based on various parameters.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 3 is a block diagram illustrating a speech recognition apparatus according to an embodiment of the present invention, FIG. 4 is a diagram for describing a structure of speech recognition data having directionality, and FIG. 5 is a general speech recognition data structure and the present embodiment. It is a figure for comparing and explaining the speech recognition data structure by this invention.

Referring to FIG. 3, the voice recognition apparatus 100 according to an embodiment of the present invention may include a voice receiver 110, a feature extractor 120, a pattern storage unit 130, a command code storage unit 140, and a controller. And 150.

In detail, the voice receiver 110 includes a plurality of acoustic sensors, for example, a sensor 1 112, a sensor 2 114, and a sensor 3 116 to receive an analog voice signal input from the outside.

In this case, the voice receiver 110 generally receives information on the size of an analog voice signal input. In addition, the voice receiver 110 is disposed in a three-dimensional space as shown in point A shown in FIG. 4, and each of the sensors 112, 114, and 116 included in the voice receiver 110 is voiced in a three-dimensional space. Receive the direction information to which the signal is input.

For example, sensors 112, 114, and 116 are respectively disposed on the x-axis, y-axis, and z-axis in three-dimensional space, respectively, and at the output location (point B or point C) of the analog speech signal. Direction information ((x1, y1, z1) or (x2, y2, z2)) is received.

In addition, the voice receiver 110 converts the voice recognition data having the main parameter for identifying the analog voice signal, for example, the magnitude information of the analog voice signal, and the direction information of the sub parameter, for example, the analog voice signal.

As illustrated in FIG. 5, the converted voice recognition data includes only size information of an analog voice signal according to a time change (T1, ..., TN) in the general voice recognition method described with reference to FIG. 1 byte of speech recognition data, but in the present invention, for example, 4 bytes of speech recognition data including the magnitude and direction information of the analog speech signal according to the time change (T1, ..., TN) .

The feature extractor 120 receives the speech recognition data converted by the speech receiver 110 and extracts a recognition feature for identifying the speech recognition data. In this case, the feature extractor 120 extracts the recognition feature for the main parameter and extracts the recognition feature for the subparameter. For example, the feature extractor 120 extracts the recognition features for each direction (x, y, z) together with the recognition features for the size of the voice recognition data.

The pattern storage unit 130 stores a predefined pattern (hereinafter, referred to as a voice model data pattern) of voice model data stored by a user in a speaker dependent system, and the common features of voice signals of an unspecified majority speaker in a speaker independent system. Model and save the registered pattern. In this case, the pattern storage unit 130 may store the registered pattern based on the main parameter, that is, the direction information. To this end, the pattern storage unit 130 may include a plurality of voice DBs 132, 134, and 136 in which a registered pattern is stored based on each direction information.

The command code storage unit 140 stores a command code corresponding to the analog voice signal input by the user. For example, a predetermined command code may be stored to stop an operation performed by a system on an analog voice signal of a user's “stop”.

In addition, the command code storage unit 140 may store an American Standard Code for Information Interchange or the like to display an analog voice signal input by a user as a character corresponding to a predetermined display area of the electronic device.

That is, the terminology of the command code used herein refers to various kinds of data in which codes for instructions for causing the electronic device to perform only a specific function.

The controller 150 receives a recognition feature of the speech recognition data extracted by the feature extractor 120, compares the patterns stored in the pattern storage unit 130, and determines whether a matching pattern exists.

Here, in the preferred embodiment, it is determined whether there is a pattern matching the recognition feature of the speech recognition data, but if the pattern actually does not exist, the most similar pattern may be extracted.

If there is a matching pattern, the controller 150 reads the command code corresponding to the matching pattern from the command code storage 140 and outputs it as a recognition result. Accordingly, the system performs the corresponding operation. In addition, when the system does not output a recognition result to perform a specific function corresponding to the analog voice signal, for example, when the character corresponding to the analog voice signal is displayed in a predetermined display area, the corresponding code is not a command code. Information about characters can also be output as a recognition result.

The voice recognition method using the voice recognition device will now be described in detail.

6 is a flowchart illustrating a speech recognition method according to an embodiment of the present invention.

4 to 6, in a voice recognition method according to an embodiment of the present invention, a step of receiving a voice signal (S100) and a step of converting the voice signal into voice recognition data having main and sub parameters (S110). , Noise processing step (S120), extracting a recognition feature of the speech recognition data (S130), comparing the recognition feature of the speech model data pattern and the speech recognition data (S140), determining whether a matching pattern exists In step S150 and outputting the recognition result (S160).

Specifically, in step S100, the voice receiver 110 receives an analog voice signal input from the outside. At this time, the voice receiver 110 receives the main parameter and the sub-parameter of the analog voice signal, for example, size information and direction information.

In step S110, the voice receiver 110 converts the voice recognition data into, for example, four bytes of voice recognition data having the main parameter and the sub parameter as shown in FIG. 5 based on the received main parameter and the sub parameter.

In operation S120, the voice receiver 110 is included in the converted voice recognition data, and separates the voice recognition data for the noise component input from the surrounding environment from the voice recognition data for the voice signal input from the real user. Remove the output.

In operation S130, the feature extractor 120 receives voice recognition data from which the noise component output from the voice receiver 110 is removed and extracts a part of the voice recognition data, for example, a collection of phonemes. In this case, the feature extractor 120 extracts the feature considering both the main parameter and the sub parameter.

In operation S140, the controller 150 compares the voice recognition model data patterns pre-stored with the extracted recognition feature of the voice recognition data based on the main parameter and the sub parameter in the pattern storage unit 130. At this time, the pattern storage unit 130 compares the pre-stored voice model data pattern and the recognition feature based on the direction information in each voice DB.

In step S150, the controller 150 determines whether a matching pattern exists as a result of the comparison in step S140. As a result of the determination, if there is a matching pattern, step S160 is performed. If the determination result does not exist, the voice recognition process is terminated. In operation S150, only the presence or absence of a matching pattern is determined for convenience of description. However, when there is no matching pattern, the most similar pattern may be recognized as a matching pattern.

In operation S160, when the matching pattern exists or recognizes the most similar pattern as the matching pattern, the controller 150 outputs a recognition result corresponding thereto.

Here, the recognition result may be, for example, when the analog voice signal of the user is a command for the system to perform a predetermined function, the command code storage unit 140 may be read out and output the corresponding command code, the predetermined function is In the case of a command to perform a function of displaying a simple character, the command code storage 140 or an ASCII code for a predefined character may be output.

According to such a speech recognition apparatus and a speech recognition method, when the recognition rate in the case of performing speech recognition is generally 1/10 considering only the magnitude of the speech signal, the x-axis and the y-axis in the directionality, that is, three-dimensional space When speech recognition is performed in consideration of the directionality of each of the z-axis, a recognition rate of 1 / (10 × 10 × 10) is obtained by comparing the predefined speech model data patterns in consideration of each directionality and performing speech recognition. And speech recognition with high precision is possible.

As described above, according to the present invention, recognition rate of speech recognition can be improved by comparing not only the magnitude of the speech signal but also various models previously defined in consideration of the directionality in consideration of the directionality in the three-dimensional space.

While the foregoing has been described with reference to preferred embodiments of the invention, those skilled in the art will be able to make various modifications and changes to the invention without departing from the spirit and scope of the invention as set forth in the claims below. It will be appreciated.

Claims (7)

A voice receiver for receiving a voice signal and converting the voice signal into voice recognition data having a main parameter and a sub parameter for identifying the voice signal; A pattern storage unit for storing a predefined voice model data pattern; A feature extractor which extracts a recognition feature of the speech recognition data according to the main and sub parameters; And And a controller configured to compare the recognition feature with the speech model data pattern and to output a command code corresponding to the speech model data pattern if a speech model data pattern matching the recognition feature exists. Speech recognition device. The method of claim 1, The main parameter is magnitude information of the voice signal, and the sub parameter is direction information on an output direction of the voice signal. The method of claim 2, wherein the voice receiver, And a plurality of acoustic sensors for identifying the direction information in a three-dimensional space region and including the direction information in the speech recognition data. The method of claim 1, And a command code storage unit for storing the command code. The method of claim 1, wherein the pattern storage unit, And a pre-defined voice model data pattern for the voice signal according to the main parameter and the sub parameter. Receiving a speech signal and converting the speech signal into speech recognition data having a main parameter and a sub-parameter for identifying the speech signal; Extracting recognition features of the converted speech recognition data; Comparing the recognition feature with a predefined speech model data pattern; And And if it is determined that the speech model data pattern that matches the recognition feature exists, outputting a command code corresponding to the speech model data pattern. The voice recognition method of claim 6, wherein the main parameter is magnitude information of the voice signal, and the sub parameter is direction information on an output direction of the voice signal.

Images