How to Write Scripts That Sound Natural in Text-to-Speech

A reader can glance back at a complicated sentence. A listener cannot. Text that reads well on a page can become hard to follow when spoken, and a synthetic voice makes that worse because it cannot add the emphasis a human narrator would. Aim for text that you could comfortably read aloud yourself in one breath per sentence.

• Prefer sentences of about 10 to 20 words. Split anything that needs more than one comma-separated clause in the middle.
• Put the main point first. "Press Save to keep your changes" is easier to follow than "To keep your changes, which you will otherwise lose, press Save."
• Avoid stacking parentheses and dashes. Turn asides into their own sentence.
• Use paragraphs. A new paragraph gives the voice, and the listener, a natural reset.

Punctuation is the main way you tell a voice where to pause and how to shape a sentence. A comma gives a short break, a full stop a longer one, and a question mark usually makes the voice rise at the end. Missing punctuation is the most common reason sentences run together.

On Open Text to Speech you can also add pause tags such as [pause:short] or [pause:long] where you want a deliberate break, for example before revealing an answer or between list items.

Voices guess how to read numbers and symbols, and the guess depends on context. When it matters, write it out:

• Acronyms. If it should be spelled out, add spaces or full stops: "U N" or "U.N.". If it should be said as a word (like NASA), leave it as one word.
• Unusual names. Spell them the way they sound for the listener, even if it looks odd on screen. Keep a short list of your phonetic spellings so you use the same one every time.
• Homographs. Words like "read", "lead", "live" and "record" change pronunciation with meaning. If the voice picks the wrong one, rephrase the sentence ("I have read" instead of "I read") so only one reading makes sense.
• Foreign words. A voice for one language may mispronounce a phrase from another. Consider a simpler synonym or a short explanation.

Each request in the Studio can contain up to 2,000 characters, roughly 300 words. For longer scripts, split at natural breaks such as paragraphs or scene changes, never mid-sentence, and keep the same voice and settings for every part so the pieces join smoothly.

• Read the script aloud yourself once. Anywhere you stumble, the voice will too.
• Check every number, date, symbol and abbreviation.
• Check names and acronyms, and use your phonetic spellings.
• Make sure every sentence ends with punctuation.
• Generate a short test passage first, then the full script.