API Playground

Disable SPLIT/MERGE detection. Compare row N to row N strictly by position.

Include character-level inline diffs for MODIFIED rows.

Detect when one original row was split into multiple reworked rows.

Detect when multiple original rows were merged into one reworked row.

Compute Character Error Rate for MODIFIED rows.

Compute CER on the transcript column only. Gives a more focused view of transcription quality.

Compute Word Error Rate for MODIFIED rows.

Compute WER on the transcript column only. Gives a more focused view of transcription quality.

Compute Segmentation Error Rate — measures how many boundary-changing events (SPLIT/MERGE/ADDED/DELETED) occurred relative to comparable rows.

Compute Sentence Error Rate — fraction of comparable (UNCHANGED + MODIFIED) rows that contain at least one change.

Compute transcript-level SER — treats each sentence in the transcript column as a unit and counts how many changed. Analogous to how transcriptWER works at word level.

Compute Speaker Attribution Change Rate — fraction of MODIFIED rows where the speaker field changed. Only reported when a speaker column is detected.

Compute the composite quality score — a weighted average of the enabled CER, WER, SegER, and SER metrics.

Normalize Arabic diacritics before comparison. Prevents inflated MODIFIED counts from harakat changes.

Match rows by index only. Ignores similarity. Only use if no rows were reordered.