KR101960434B1 - Tagging method in audio file for machine learning - Google Patents

Abstract

The present invention relates to a method for executing tagging in a voice file and a tagging apparatus in which software for executing the method is installed.
The method of the present invention includes a tagging player user interface capable of executing a process for machine learning by refining voice data and executing a tagging player and performing an operating event on a voice file stored in a storage or generated in real time Selecting a specific section for the voice file and inputting the text tagging.

Description

TECHNICAL FIELD The present invention relates to a tagging method for tagging a voice file,

The present invention relates to a speech / sound tagging technique for machine learning. Particularly, the present invention has been completed as part of a national research project to develop situation judgment solution technology based on voice / sound analysis.

In order for the intelligent robot to effectively judge the situation, it must learn the collected data in advance or in real time. Today's human activities in the network generate a lot of information and data that are difficult to quantify. They are gathered back over the network to create a big data environment. In a big data environment, you must include unstructured data, not just structured data. However, since unstructured data also includes audio files, a method of extracting meaningful data from audio files has been studied.

For example, consider a customer consultation voice file in a shopping mall. If the words appearing in the voice files can be tagged and stored in advance, the stored data can easily grasp the products that the customers are most interested in or the inconveniences of the customers. Another example is to think about the voice file of a report by a public agency. Similarly, various information can be obtained by gathering inconveniences or accident information of residents through words appearing in voice files.

However, in order to achieve such a task, the collected voice data had to be refined into machine learning objects. This is the prior art for extracting data from speech files into machine learning data. After the system administrator listens to the voice file (first process), the specific information and the time information in which the information appeared appear separately and the data is manually stored (second process). This is a disadvantage in that a mistake of the manager occurs and the working time is slowed down.

The inventors of the present invention have studied and developed for a long time in order to solve such a problem, and finally the present invention has been completed.

The present invention proposes a technique for making voice data more adaptable to a big data environment. In particular, an object of the present invention is to provide a method for directly tagging a specific section by a player who reads and plays an audio file, in order to make the audio file, which is atypical data, easy to machine learning. This minimizes the likelihood of user error during tagging and storage, quickly and easily.

On the other hand, other unspecified purposes of the present invention will be further considered within the scope of the following detailed description and easily deduced from the effects thereof.

The present invention relates to a tagging user interface method for machine learning that performs tagging to a voice file,

A tagging player user interface that can execute a tagging player as a process execution for machine learning by purifying voice data and can perform an operating event for a voice file stored in a storage or generated in real time, And inputting text tagging.

The tagging user interface method for machine learning which performs tagging to a voice file according to a preferred embodiment of the present invention is characterized in that the operating event is one of a voice file retrieval, a reproduction, a reversion, a skip, Button is selected.

The tagging user interface method for tagging a voice file according to a preferred embodiment of the present invention further includes generating and storing a voice file into which the text tagging is inputted as a raw data file for machine learning can do.

In addition, in the tagging method for machine learning for performing tagging in a voice file according to a preferred embodiment of the present invention, the text tagging may include first classification data in a first input UI, second classification data in a second input UI, It may be to input data.

In addition, the tagging user interface method for machine learning that performs tagging on a voice file according to a preferred embodiment of the present invention may further include a step of learning module for learning the raw data file.

The above-described problem of the present invention can solve the conventional problem that voice file execution and tagging work are uncomfortable. This is because a complicated method of refining speech data for existing machine learning can be easily performed through a user interface for a voice file player. In addition, since the intermediate process is omitted, it is possible to prevent a user from making a mistake and to perform the task in a short period of time.

On the other hand, even if the effects are not explicitly mentioned here, the effect described in the following specification, which is expected by the technical features of the present invention, and its potential effects are treated as described in the specification of the present invention.

1 is a diagram schematically showing an example of an electronic configuration of the tagging device of the present invention.
FIG. 2 is a diagram showing an example of a user interface configuration of the tagging player 100 according to a preferred embodiment of the present invention.
FIG. 3 shows an example of a raw data Json Format stored after executing the tagging process according to the present invention.
Figure 4 schematically illustrates the overall process of a method according to one embodiment of the present invention.
* The accompanying drawings illustrate examples of the present invention in order to facilitate understanding of the technical idea of the present invention, and thus the scope of the present invention is not limited thereto.

In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may obscure the subject matter of the present invention.

1 schematically shows an example of the configuration of a tagging apparatus according to the present invention.

The tagging device of the present invention consists of an organic combination of hardware and software. The tagging device accesses the first audio file 150 stored in the storage 15 and performs a tagging operation. Then, the tagging device stores the second audio file 151 converted for data collection.

In one embodiment of the present invention, the repository 15 is a repository that resides on a network. It can be a repository that can be accessed by a tagging device by a company intranet, or it can also be a repository that can be accessed by a tagging device via the Internet. In other words, the voice files 150 and 151 become data to be accessed, spread, shared, and analyzed through the network. In this embodiment, the tagging device therefore has a communication module capable of network communication.

Also in another embodiment of the present invention, the storage 15 comprises internal or external memory of the device. It is also possible to tag an audio file stored in the device memory by tagging or generate a voice file on the device.

The input device 13 may include input means such as a mouse, a touch input device, and a keyboard. The input device 13 executes an input operation required for an operating event or a tagging event in the tagging player 100. The display 14 outputs the user screen of the tagging player 100 of the tagging device.

The control unit 10 includes one or more processors and executes the computer code together with the operating system to perform operations to generate and use data. The control unit 10 can also receive and process input and output data between the components of the tagging device using a series of commands. Particularly the first audio file 150 in the repository 15 through the execution of the tagging player 100 to control a series of processes for executing an operating event and / or a tagging event.

Although the tagging player 100 is shown as belonging to the control unit 10 for convenience of explanation, the tagging player 100 is a software module stored in the memory of the tagging apparatus. Such a memory is not shown in FIG. 1 but is a conventional means of providing a place for storing computer codes and data used in the device.

A user (a user who performs an operation to refine an audio file for machine learning) executes the tagging player 100 to retrieve or manipulate an audio file or execute a tagging operation. At this time, the microphone 11 and the speaker 12 of the tagging apparatus are executed.

2 shows an example of a user interface configuration of the tagging player 100 according to a preferred embodiment of the present invention.

The tagging player 100 includes an operating event user interface 110 and a tagging event user interface 130.

The operating event user interface 110 provides a means for executing operations on voice files in the tagging player. For example, the playback button 112 has a function of playing back a voice file, the revert button 111 has a function of returning a voice file backward, the quick playback button 116 has a function of quickly playing back a voice file, 114 stops playback of the audio file temporarily, the stop button 115 stops playback of the audio file, and the record button 113 supports the function of the user to record the sound. It can also include various voice manipulation functions such as skip and skip back. In this specification, the operation of the voice file by selecting the function button is expressed as an operating event.

As described above, the tagging player 100 includes the operating event user interface 110, which is advantageous in that it can also function as a voice file player.

The tagging event user interface 130 provides a function for the user to input and tag text using the input means. Preferably, a plurality of classification tagging operations can be executed. For example, in the first input UI 131, data on <status> may be inputted as first classification data, and data on <word> may be inputted as second classification data in the second input UI 132.

The tagging event user interface 130 includes two input UIs in the illustrated example, but may add an input UI and may be provided with a button for adding an input UI. Theoretically, the user can define up to M classifications (M is an integer greater than 1) in the first classification, and the M input UI can be provided by the tagging event user interface 130 accordingly. For each input UI, the user can perform tagging tasks by entering one or more label texts.

In the embodiment of FIG. 2, the tagging operation is performed with the label <fire> as the first classification data, and the tagging input is performed with the label <accident> as the second classification data. At this time, the user can set the interval. The user can select the start time and end time of the section selected by the user using the input means.

Preferably, a waveform proportional to the time of the audio file is output to the area 120 of the screen. In this way, it is possible to improve the efficiency of the tagging event that performs the tagging operation by setting the operating event and the tagging interval for the voice file.

In the tagging player 100 of FIG. 2, buttons for loading voice files and buttons for storing tagged files are not shown, but this should be understood as means of course included in the present invention.

In the present invention, the tagging player 100 is used as a tagging tool to select a specific section of a voice file, and then the tagging information can be tagged by the user in the corresponding section. Then, finally, the information tagged by the user is converted into a raw data file which is easy to machine learning.

The tagging function supported by the above-described tagging tool of the present invention may include the following:

- You can create a label with the name you want.

- You can insert data by selecting the desired voice section on the label created by the user.

- It can provide correction / deletion function for data entered by the user.

- You can generate the raw data as user input data.

- You can add / modify / delete Raw Data file created by user.

On the other hand, if you export the information that you entered ^Export using the tagging player (100), Raw Data yireohada Examples simplify the rules for creating file: Create Raw Data file name that is based on the name of the sound file you The name of the Label specified by the extension is assumed. For example, if the name of the voice file is "MyLife.wav" and the label name of the data input by the tagging tool is "keyword", the name of the raw data file to be generated is "MyLife.keyword".

The data in the raw data is generated in the Json format, which can be created as shown in FIG.

LABEL_NAME is the label name added by the user, DATA is the data entered in the section selected by the user, START_TIME is the name of the first voice file, Is the start time of the section selected by the user, and END_TIME is the end time of the section selected by the user.

FIG. 4 shows a whole process of a tagging user interface method for machine learning in which tagging is performed on a voice file according to a preferred embodiment of the present invention, whereby the subject matter of the present invention is summarized again.

The input means of the tagging device executes the tagging player described in Fig. 2 (S100). Then, the operating event can be executed in the tagging player operating event UI (S110).

Although FIG. 4 does not show the process of loading the voice file, it is preferable that the voice file in the storage has been called before executing the operating event. Or you may be creating a voice file in real time using the recording function.

In addition, the user inputs the data of the voice file in the tagging player using the input means and executes the tagging operation (S120). At this time, as described above, the data is input by selecting a specific section as the input means during the tagging operation.

Next, the tagging device generates and stores a voice file into which the text tagging is inputted as a machine learning Raw Data file (S130).

The reason for tagging voice files with the above process is to collect big data effectively from unstructured voice files. Therefore, the step of learning the raw data file stored in the repository, which consists of software algorithms that perform machine learning, will continue with the subsequent process.

For reference, the machine learning tagging user interface method for tagging a voice file according to an embodiment of the present invention may be implemented in a form of a program command that can be executed through various computer means and recorded in a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs, DVDs, magneto-optical media such as floptical disks, A hard disk drive, a flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, as well as machine accords such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The scope of protection of the present invention is not limited to the description and the expression of the embodiments explicitly described in the foregoing. It is again to be understood that the present invention is not limited by the modifications or substitutions that are obvious to those skilled in the art.

Claims (5)

A tagging player user interface capable of executing a tagging player, which is a process execution for refining voice data for machine learning, to execute an operating event for a voice file stored in a storage or generated in real time, selects a specific section for the voice file And generating and storing a text file into which the text tagging is inputted as a machine learning learning data file, wherein the text tagging is a processing of inputting first classification data in a first input UI, Wherein the second classification data is input to the voice file. The method according to claim 1,
Wherein the operating event is executed by selecting any one of a loading, a playback, a reverting, a skip, a quick playback, and a recording of a voice file. delete delete The method according to claim 1,
Further comprising the step of the learning module learning the raw data file.

Images