Persian reading, built for accuracy.

VowelMarks reaches 96.1% word-level pronunciation accuracy and preserves the original text across every sentence in its current source-preservation test.

View benchmark results

Source text

پردازش دقیق زبان فارسی

Diacritization + Ezafe

پَردازِشِ دَقیقِ زَبانِ فارسی

Pinglish

pardaazesh-e daghigh-e zabaan-e faarsi

Translation

Accurate Persian language processing.

High accuracy across Persian pronunciation and context.

VowelMarks delivers reliable pronunciation and Ezafe support, resolves context-dependent readings, and keeps the original Persian text intact. For Persian text to speech, these results describe the reading layer that prepares pronunciation before audio is generated; they do not measure voice naturalness or claim that every recording is correct.

Pronunciation accuracy: 96.1%2,590 words across 300 Persian sentences.
Ezafe precision: 96.5%93.9% recall across 800 Persian sentences.
Ambiguous word accuracy: 90.8%491 examples of words whose pronunciation changes with context.
Source text preserved: 100%No changes to letters, punctuation, or spacing across 337,107 sentences.

Coverage and source preservation

Broad Persian coverage. Original text preserved.

The test collection includes historical news, current Persian publishing, formal prose, colloquial writing, mixed Persian and Latin text, and deliberately difficult examples.

337,107unique Persian sentences tested

263,790: sentences containing a word with more than one possible pronunciation
319,754: sentences containing a possible Ezafe construction
20,131: sentences combining Persian and Latin script
8,364: sentences from current Persian news sources

In every source-preservation check, VowelMarks retained the original letters, punctuation, and spacing. Any marked result that changes the submitted text is discarded.

How accuracy is measured

How each result is measured.

Pronunciation, Ezafe, words with multiple readings, and source preservation are measured separately because each requires a different test.

AreaTestResult

Pronunciation300 Persian sentences, 2,590 words96.1% word-level accuracy

Ezafe detection800 Persian sentences96.5% precision, 93.9% recall

Words with multiple readings491 context-dependent examples90.8% contextual accuracy

Source preservation337,107 unique Persian sentences100% of source text preserved

Evaluate VowelMarks

Bring the Persian your product actually uses.

We can review representative text from your workflow and help you evaluate the capabilities your product needs.

Request evaluation details Explore the API