persian-tts-preprocessor

name	Persian TTS Preprocessor
description	Use when preparing Persian / Farsi text for any TTS engine. Returns the same text with normalisation, half-spaces, punctuation, Ezafe marks, exception fixes, and optional spoken-style rewrite.
dependencies	[]
category	productivity
subcategory	writing-assist
tags	["persian","farsi","tts","text-to-speech","preprocessing","normalization","ezafe","ssml"]
author	{"name":"Nima Aksoy","url":"https://nimaaksoy.com","github":"nimaaksoy"}
license	CC-BY-4.0
version	0.3.0
created	2026-05-17T00:00:00.000Z
updated	2026-05-17T00:00:00.000Z

Persian TTS Preprocessor

Overview

Persian TTS engines (Azure, Google, MMS, XTTS, ElevenLabs) usually fail not because the model is bad, but because the input text is ambiguous. Persian doesn't write short vowels, often omits the ezafe (ـِ), uses elision in spoken form, and is sensitive to punctuation and syntax for prosody. Hand a raw paragraph to any engine and you'll get a stiff, book-like read with mispronunciations.

This skill takes a Persian text in, applies the minimum set of changes that make it read correctly, and returns the same text — same content, same paragraph structure, same line breaks — just TTS-ready.

When to use this skill

• The user has Persian text and wants to feed it to a TTS engine (Azure, Google Gemini-TTS, MMS-TTS, XTTS, ElevenLabs, Coqui).
• The user is producing voiceovers, audiobooks, podcasts, IVR, or accessibility output in Persian.
• The user's previous TTS output was stiff, mispronounced, run-on, or read numbers as digits.
• The user has text that came from ASR / OCR / LLM and needs punctuation restored before TTS.

Do not use when:

• The user wants to translate Persian to another language — that's a translation skill.
• The user wants to write new Persian text (a poem, a song, marketing copy) — that's a writing skill.
• The text is non-Persian — apply a language-specific preprocessor.

Tag legend

• [Rule] — always apply; safe in every register.
• [Conditional] — apply only when the input contains the trigger pattern.
• [Register-dependent] — apply only when the user asks for the Fluent-formal or Spoken register.
• [Engine-specific] — apply only when the target engine is known to need it.

Instructions

Step 0 — Pick the register

Ask the user once, then hold it for the whole text:

Register	When to use	What changes
Preserve (default)	The user wants minimal interference — formal writing, news, audiobooks of literary work, legal/academic text	Only safe, mechanical fixes
Fluent-formal	The user wants formal but smooth reading — explainer videos, professional voiceover, e-learning narration