Full Page Transcription (FPT) captures the visible text on a page instead of extracting only predefined fields. Use FPT for documents with Unstructured layouts when you need page-level text and location data for downstream processing.
FPT identifies and transcribes page segments without matching the page to a predefined layout. This approach supports documents whose content or structure varies between submissions.
How Full Page Transcription works
FPT processes each page independently and adds the transcription results to the page output.
Page selection: The Full Page Transcription block receives a Submission or a list of pages. For a submission, the block processes its unassigned pages.
Segmentation: The system divides each page into text, checkbox, and signature segments. Each Segment includes a Bounding Box that identifies its location on the page.
Transcription: The system processes each segment and records its content. Text segments contain transcribed text. Checkbox and signature segments contain their detected state.
Output: The system attaches the segments to their source pages. Downstream blocks can use the text, segment type, location, and confidence information.
By default, FPT skips pages that already contain segment results. Turn on Rerun Segmentation to process those pages again.
When to use Full Page Transcription
Choose FPT according to the output that your business process requires.
Use FPT when | Use field-based extraction when |
|---|---|
You need the visible text from most or all of a page. | You need specific business values, such as an invoice number or total. |
Documents do not follow a consistent layout. | Documents match Structured or Semi-structured layouts. |
A downstream process needs page text and segment locations. | A downstream process needs named fields and table cells. |
FPT produces page content, not a business data schema. If a downstream system requires named fields, add logic that transforms the FPT output or use a field-based extraction flow.
Full Page Transcription output
The output contains the processed pages and their segments. A segment can include the following information:
A segment identifier
The segment type: text, checkbox, or signature
A bounding box that locates the segment on the page
The transcription or detected state
Segmentation and transcription confidence information
The block can return a submission object or a list of page objects. In both modes, the segment results remain associated with their source pages.
Full Page Transcription QA
Full Page Transcription QA measures the Accuracy of FPT results. The system samples a configured percentage of segments and creates tasks for human review.
During a QA task, a reviewer compares each highlighted segment with its transcription. The reviewer marks the segment or token as Correct, Incorrect, or Illegible.
A correct result counts as a correct machine read.
An incorrect result counts as an incorrect machine read.
An illegible result is excluded from QA reports.
An invalid page is abandoned and excluded from QA reports.
You can split text segments into smaller tokens for QA. Token splitting can make long segments easier to review. A segment counts as correct only when all its reviewed tokens are correct.
FPT QA evaluates transcription accuracy. It does not replace the transcription with a corrected value. Add a separate Supervision step when a keyer must correct low-confidence or incorrect text.
Language selection
Configure the language family that matches the document content. If the flow does not provide a language family, the underlying FPT process uses Latin.
The selected language family also determines the reading order used in QA. For example, the system sorts segments in right-to-left order for supported right-to-left languages. To learn more about languages, see Supported Languages.
Full Page Transcription reporting
Use FPT reporting data to measure machine accuracy, review volume, and processing time.
Accuracy: Compare machine and manual segment accuracy. Filter the data by flow or segment type.
Supervision volume: Track the number of segment transcriptions completed during FPT QA.
User performance: Review the time, segments, and characters completed by each reviewer.
Machine working time: Measure how long FPT work waits for processing and how long processing takes.
Requirements and limitations
FPT identifies page segments but does not assign business meanings or field names to them.
Transcription quality depends on the source image and the applicable models. Use FPT QA to measure accuracy with representative documents.
FPT QA reviews only pages that contain segment results.
Turning off Rerun Segmentation preserves existing segment results and processes only pages without them.