- Print
- PDF
Long sentence recognition builder
- Print
- PDF
Available in Classic and VPC
CLOVA Speech's builder allows you to request and manage voice recognition tasks with a UI environment. A builder is provided for each domain, and the following menus can be used in the builder.
- Task list menu: register and manage recognition tasks, and edit and export task results
- Keyword boosting menu: register words whose recognition probability you want to increase as keywords and manage registered keywords
- Setting menu: check domain information
Depending on domain types, some features may not be exposed.
Execute builder
To execute a builder, follow these steps:
To use a builder, you must first create a domain. For more information on domain creation, see Create domain.
- Click the environment you are using in the Region menu and the Platform menu on the NAVER Cloud Platform console.
- Click Services > AI Services > CLOVA Speech in order.
- Click the Domain menu of the Long sentence recognition type.
- Click the [Execute builder] button of the domain.
Request and manage recognition tasks
You can request recognition tasks, and check the progress of each task in the builder's Task list menu. Also, you can use the recognition result editor to edit recognition results and export them in the format of your choice.
Task list screen
The basic description of the Task list menu for using the builder is as follows:
Field | Description |
---|---|
① Menu name | Name of the menu currently being viewed |
② Task management function | Requesting, deleting, and exporting and re-recognizing recognition tasks |
③ Search bar | Searching tasks by name of the task target file or location of the result file |
④ Task list | List of recognition tasks being searched |
Check recognition task list
To check information on recognition tasks registered in the builder, follow these steps:
- Click the Task list menu from the builder of the domain.
- When the domain list appears, check the information on the recognition task.
Task target: name of the recognition task target media file
Language: choose the language to use for the recognition task
Task status: the current status of the task, indicated as one of the following
Task status Description Task pending Waiting for a recognition task - All servers are working, but when a server becomes available, it changes to Task in progress
- [Cancel recognition task] button: click it to cancel the task
Task in progress The recognition task is in progress Cancellation completed Cancellation of the recognition task is completed Canceling The recognition task is being canceled Task failed The recognition task failed - : click to check the reason for task failure
Task completed The recognition task is completed Request method: how to upload a task file
- To Select from Object Storage, you can click to go to the Object Storage console screen
Result file location: Object Storage path where result files are saved
- You can click to go to the Object Storage console screen
Task start date (UTC+09:00): the date and time the task was started
Task end date (UTC+09:00): the date and time the task ended
Edit recognition results: click the [Edit recognition result] button to execute the recognition result editor. (See Using the recognition result editor.)
Request recognition task
To request a recognition task from the builder, follow these steps:
- Click the Task list menu from the builder of the domain.
- Click the [Request recognition task] button.
- Click the appropriate tab for the method to upload the recognition target file.
- [Select from Object Storage] tab: select a file from the domain's recognition target storage path and sub-path
- [Upload File] tab: select a file in the local PC
- Enter task information.
- Result file location: select the Object Storage path to save the result file
- The domain result file storage path and sub-path can be selected
- Language settings: select the language to use for the recognition task
- Use speaker separation: select whether to use speaker separation
- Detect event: select whether to enable detection of events (applause, laughter, music) in the sound source
- Keyword boosting: select whether or not to use keyword boosting. (For more information on keyword boosting, see See keyword boosting)
- Result file location: select the Object Storage path to save the result file
- Upload the recognition target file.
- If you selected the [Select from Object Storage] tab: select a file from the Select target file field
- You can double-click a folder to check the files in the folder.
- If you selected the [Upload file] tab: drag the file with your mouse or click here. Drag the file into the field, or click Drag the file with the mouse, or click here to select the file
- If you selected the [Select from Object Storage] tab: select a file from the Select target file field
- Click the [Begin request] button.
- When the task is completed, you can check the recognition result file (JSON) in the path of Result file location
The maximum number of speakers that can be recognized from one sound source is 10.
If you're using the re-recognition function, you don't need to send a recognition task request again for the same sound source.
Use recognition result editor
When the recognition task is completed, you can run the recognition result editor to check the recognition target media file and recognized texts together. Also, you can edit recognition results, and export them in the desired format.
Recognition result editing screen
When the recognition result editor is executed, the recognition result editing screen appears. The basic description of the recognition result editing screen is as follows:
Field | Description |
---|---|
① Export | Exporting edited results |
② Recognition target file |
|
③ Filter | Filter recognition results based on the speaker and reliability |
④ Voice recognition result |
|
⑤ Event detection result |
|
⑥ Playback | Playback recognition target files and control their volume |
Edit and export recognition results
To edit recognition results in the recognition result editor and export them in the desired format, follow these steps:
You can use Export recognition result to export the result files of multiple tasks at once.
- Click the Task list menu from the builder of the domain.
- Click the [Edit recognition result] button for completed tasks in the task list.
- Check the recognition result in the recognition result editing screen.
- When a media file is played back, the text corresponding to the voice currently played back is highlighted in the timeline
- [Speaker name] button: click it to filter recognized texts by speaker
- [Recognition result reliability] button: click it to filter recognized texts by reliability of recognition results
- If necessary, edit recognition results.
- Edit speaker: click the drop-down menu to edit the speaker of the text
- : click it to delete a speaker
- : click it to edit the name of the speaker
- [Add speaker] button: click it to add a speaker
- Edit text: click the row marked as to edit the contents, and click .
- To cancel editing, click .
- Edited contents are highlighted in green.
- Edit speaker: click the drop-down menu to edit the speaker of the text
- To export recognition results, click the [Export] button, enter settings information and click the [OK] button.
- File name: enter the name of the result file to export
- File format: click the drop-down menu to select the format of the file to export (select json, smi, srt, txt, csv or xls)
- Push back timeline: push back the timeline of recognition results by the set time
- Export method: select Object Storage or Download
- If Object Storage is selected, click File location to select the path for storing the recognition result file
- If Download is selected, download the recognition result file to the local PC
Export Recognition Result
To export recognition result files for multiple tasks, follow these steps:
You can export only the recognition result files of tasks whose Task status is Task completed.
- Click the Task list menu from the builder of the domain.
- Click and select the task to export.
- You can select multiple tasks simultaneously.
- Click the [Export] button.
- When the Export popup window appears, check the export settings, and click the [OK] button.
- : click it to edit the name of the recognition result file
- : click it to delete it from the export list.
- File format: click the drop-down menu to select the format of the file to export (select json, smi, srt, txt, csv or xls)
- Export method: select Object Storage or Download
- If Object Storage is selected, click Result file location to select the path for storing recognition result files.
- If Download is selected, download the recognition result file to the local PC
Delete recognition task
To delete registered recognition tasks, follow these steps:
- Click the Task list menu from the builder of the domain.
- Click to select the task to be deleted.
- You can select multiple tasks simultaneously.
- Click the [Delete] button.
- When the Delete task pop-up window appears, click the [Delete] button.
Keyword boosting
You can increase the recognition rate for a specific word through keyword boosting. The Keyword boosting menu allows you to register and manage keywords.
Keyword Boosting screen
The basic description of the Keyword boosting menu for using the builder is as follows:
Field | Description |
---|---|
① Menu name | Name of the menu currently being viewed |
② Keyword management function | Upload keyword files, and download and delete registered keywords |
③ Search bar | Search keyword |
④ Keyword list | List of keywords being searched |
Add keywords
To add keywords for an increased recognition rate, follow these steps:
- You can register up to 1000 words, and Korean and English are supported.
- It is recommended to register fewer than 500 keywords. Performance may deteriorate if more than 500 words are registered.
- More similar words are recognized with higher weights, and this can affect recognition results.
You can register keywords in any of the following ways:
- Upload keyword files: add multiple keywords at once by uploading an Excel file containing keywords
- Add individual keywords: enter keywords in the builder and add them
Upload keyword file
To upload an Excel file containing keywords and register multiple items at once, follow these steps:
- Create an Excel file containing keywords and weights.
- Enter an integer between 1 and 5 for the weights.
- If you click the Upload template in the screen for procedure No. 4, you can check the template of the keyword file.
- Click the Keyword boosting menu in the builder of the domain.
- Click the [Upload] button.
- When the Upload popup window appears, drag the file with the mouse or click here. Drag the keyword file in the local PC into the field, or drag the file with the mouse or click here. Click the field and select a keyword file.
- Click the [OK] button.
- Overwrite: if you click the checkbox to select it, all existing keywords are deleted and new keywords are registered
If the keyword file contains keywords that overlap with existing keywords, keyword registration will fail. Delete overlapping keywords or click the Overwrite check box when uploading a file.
Add individual keywords
To directly add keywords in the builder, follow these steps:
- Click the Keyword boosting menu in the builder of the domain.
- Enter the keywords to add. Click the field and enter keywords.
- Select a weight in the Weight drop-down menu.
- Click the [Add] button.
Edit keywords
To edit registered keywords, follow these steps:
- Click the Keyword boosting menu in the builder of the domain.
- From the keyword list, click of the keyword to edit.
- Click the keyword input field to edit the keyword, and click .
- To edit a keyword, click .
- To edit a weight, select a weight in the Weight drop-down menu.
Download keywords
To download registered keywords at once, follow these steps:
- Click the Keyword boosting menu in the builder of the domain.
- Click the checkbox of the keyword to be downloaded from the keyword list to select it.
- You can select multiple keywords simultaneously.
- Click the [Download] button.
- The Excel file including keywords and weight information is downloaded.
Delete keywords
To delete the registered keywords, follow these steps:
- Click the Keyword boosting menu in the builder of the domain.
- Click the checkbox of the keyword to be deleted from the keyword list to select it.
- You can select multiple keywords simultaneously.
- You may also click the [Delete] button of the keyword to immediately delete the keyword.
- Click the [Delete keyword] button.
- When the alert pop-up window appears, click the [OK] button.
Check API call information and domain information
The Settings menu allows you to check link information for API calls and information on the domains connected to the builder.
Check API call information
To check the link information to call an API, follow these steps:
You can also call the API to send files to be recognized and receive recognition results. For information on how to call the API, see CLOVA Speech API guide.
- In the builder of the domain, click the Settings menu.
- Click the [Link information] tab.
- Check the Secret Key and CLOVA Speech Invoke URL of the domain.
- [Create] button: click it to create/re-create the secret key
- [Copy] button: click it to copy the information to the clipboard
Check domain information
To check the information on domains connected to the builder, follow these steps:
- In the builder of the domain, click the Settings menu.
- Click the [Domain information] tab.
- Check the information on domains connected to the builder.
- Domain information
- Domain name: the name of the domain
- Domain code: the code of the domain
- Type: the pricing plan applied to the domain
- Storage settings
- Result file storage path: object storage path for storing recognition result files
- Recognition target storage path: object storage path for selecting recognition target files
- [Shortcut to Object Storage] button: click it to go to the console screen of Object Storage
- Domain information