Use the same script, then judge two different things
Give each voice the same sixteen sentences. First check the words: vowels, Ezafe, names, numbers and anything missing. Then listen for pace, pauses, expression and consistency. Keep the two sets of notes separate so you can tell a reading error from a delivery you simply dislike.
Copy only the Persian sentences into the voice tool
Use the unmarked text first. It asks the voice to resolve ordinary Persian spelling from context, rather than handing it all the vowels in advance. Keep the sentence pairs together. If you test marked input later, label it as a separate run instead of comparing it silently with another voice's unmarked recording.
کتاب جدید نرگس روی میز است. خانهٔ کوچک ما نزدیک مدرسه است. گل قرمز را در گلدان گذاشت. باران بارید و کفشها پر از گل شد. آن مرد در ایستگاه منتظر بود. پرندهٔ زخمی دیشب مرد. صبح یک لیوان شیر نوشید. شیر باغوحش کنار درخت خوابید. نرگس و بهرام از تهران به شیراز رفتند. جلسه ساعت ۸:۳۰ شروع میشود. قیمت کتاب ۱۲۵ هزار تومان است. شمارهٔ اتاق ۲۰۴ است. فایل PDF را با Wi-Fi فرستادم. امروز نمیروم؛ فردا میآیم. مامان، کتابت رو روی میز گذاشتم. گفت: «هنوز آماده نیستم.» بعد در را بست.
Download the Persian test script and the listening worksheet. The worksheet includes the intended readings; do not put its English notes in the narration box.
Choose standard Iranian Persian for this run. Keep the voice, pace and direction fixed, and write down the tool, settings and date. If you compare a calm lesson voice with a dramatic storyteller, you are comparing two uses as well as two voices.
The answer key: what each sentence is testing
1–2: Ezafe connects a phrase, not every adjacent word
In the first line, listen for ketab-e jadid-e Narges: “Narges's new book.” The second has khane-ye kuchak-e ma: “our small house.” The connecting sound belongs inside those phrases. In particular, the final ast should not acquire a linking vowel simply because another sentence follows.UT Austin's Ezafe lesson explains these links and the -ye form.
3–6: the sentence has to choose the short vowel
گل: gol or gel
The vase sentence calls for gol, flower, with gol-e ghermez, “red flower.” The rain-and-shoes sentence calls for gel, mud. A smooth voice that says “flower” in both has failed this pair.
مرد: mard or mord
The person waiting at the station is mard, a man. The injured bird mord, died. The written letters match; the intended vowel does not.
Keep these pairs as complete sentences: the vase, muddy shoes and injured bird supply the clues to the reading. For more examples, use the Persian homograph list and chart.
7–8: different meanings do not always require different sounds
The glass contains milk; the zoo contains a lion. Both uses ofشیر can be shir. This pair is a check on the reviewer, too. Do not mark a voice wrong for failing to produce a pronunciation difference that Persian does not require.
9: ordinary names, then your own project names
Listen for Narges, Bahram, Tehran and Shiraz. These common names give you a manageable starting point; they do not establish that the voice can pronounce every surname or regional place name. Before testing an unfamiliar person's name, obtain their preferred pronunciation or a trusted recording. Do not decide it from an English spelling alone.
10–12: numbers must retain their meaning
- ۸:۳۰: eight thirty. Hasht-o nim (“eight and a half”) or an intelligible eight-thirty clock reading is acceptable.
- ۱۲۵ هزار تومان: sad-o bist-o panj hezar toman, 125 thousand tomans. Dropping “thousand” changes the price.
- ۲۰۴: room 204. “Two hundred and four” or a clear digit-by-digit room number can work. Choose the project's convention before scoring consistency.
This is not a test of one compulsory wording for every number. The question is whether a listener hears the correct time, amount or identifier. Add a phone number or date separately if your actual app needs one, with the intended grouping written down. The number-preparation guide shows how to make those spoken forms explicit without changing the original document.
13–16: mixed script, negation, register and quotations
PDF and Wi-Fi should remain recognizable; a clear Persian-accented form and a clear English form need not count as different levels of accuracy. In line 14, the negative verbnemi-ravam must stay negative. Line 15 is deliberately casual: ketabet(“your book”) and ro belong in that register. In the quotation, check that the voice says the dialogue and the final action once, without reading punctuation names aloud.
Write down the line, the time and what you heard
Listen once without stopping, then replay the lines that sounded doubtful. Record the line number and time in the audio. For the reading, use three labels:correct, wrong or uncertain. Ask a proficient Persian listener to resolve uncertain entries; an unfamiliar accent is not automatically an error.
On a separate listen, note pace, phrase breaks, expression, voice consistency, clipped endings and omitted or repeated words. A flat but correctly read lesson and an expressive recording with the wrong price have different problems. Averaging them into one number would make it harder to choose the right voice for the job.
A useful note is specific
“Line 4, 00:12: the voice says gol where the muddy shoes require gel.”
That is actionable. “Persian sounds off” does not tell you whether to change the reading, the voice or the pacing.
If you want to compare products publicly, retain the actual files and disclose the voice, tier, settings, input and generation date. Use independent listening review and report uncertainty. This page supplies no recorded head-to-head result. The VowelMarks reading evaluation is also not a substitute for listening to generated speech.
Add three sentences that matter to your work
A vocabulary app needs reliable short examples. A teacher may care about slow, connected reading. A creator needs the same names and delivery across episodes. Add one ordinary sentence, one difficult term and one number or quotation from your own project. Keep the expected reading with the script so you can repeat the test after changing voices.
Try the script in VowelMarks Persian text to speech. Basic and Premium are separate listening options; Premium's presets and Custom Voice Studio let you test delivery as well as the reading. For an integration, use the same sentences in the Python or JavaScript API example. If you find a wrong reading, the pronunciation-repair guide helps isolate the smallest useful change.
Frequently asked questions
How can I test whether a Persian TTS voice pronounces words correctly?
Give each voice the same full sentences, identify the intended readings before listening and record errors separately from voice naturalness. Include Ezafe, words with different vowels under the same spelling, names and numbers.
Does a natural-sounding Farsi voice guarantee accurate pronunciation?
No. A fluent voice can still select the wrong short vowel, drop an Ezafe link or misread a name or number. Listen for both the reading and the delivery.
Is this a ranking of the best Persian text-to-speech tools?
No. It is an original reusable test script and listening checklist. A defensible ranking would require actual recordings, fixed settings, documented dates and independent Persian listening review.
Should different meanings of شیر sound different?
Not necessarily. Milk and lion can both be pronounced shir. A test must not invent a vowel contrast where the meanings share the same pronunciation.
Sources and review notes
Sources are listed for the claims they support. Original practice examples are identified in the article. Product capabilities and access terms were checked on the review date and can change.
- University of Texas at Austin, Ezāfe 1. Noun-phrase links and the -ye form after vowel-ending words.
- University of Texas at Austin, System of Transcription. Persian sound notation and the need to distinguish written letters from their reading.
- Rahmati and Sameti, GE2PE (2024). A research reference for evaluating Ezafe, homographs and unfamiliar words separately. Its results are not scores for this diagnostic.
- Persian With El, grammar exercises. An editorial reference for pairing a task with an answer key. All sentences and the listening worksheet here were written for this guide.