Text to speech where you pick the voice model and see the cost first

Start on Kokoro. It costs the same as the default engine and gives you six voices instead of two. The AI4Chat Voice Engine is a two-voice preset on the same Kokoro model, so there's no quality reason to stay on it once you've heard the others.

Move to Orpheus when the script has feeling in it: a short story, an ad read, a character line. It's built for intonation and emotion and costs seven times what Kokoro does, which is fair for a 300-word piece and silly for a long help article. Try Sesame CSM for lines that should sound like half of a conversation. ElevenLabs is the premium option and by far the most expensive; we'd keep it for the final take of something people will hear many times, after the wording is settled on a cheaper model.

Speech is charged in blocks of 1,000 characters, rounded up, with a minimum of one block. Spaces, line breaks and punctuation all count. The rounding is where people lose credits without noticing: a 2,050-character script costs three blocks, and trimming 51 characters brings it down to two.

Text written for the eye often sounds wrong in the ear. A few habits fix most of it:

• Punctuation is your pause control. There's no speed slider, so a full stop asks for a beat, a comma for a shorter one, and a paragraph break for the longest.
• Keep sentences short. A sentence with three clauses comes out rushed or flat on every model.
• Spell out what you want said: "forty dollars a month", "four point five stars". Models guess at digits, and not always the way you would.
• Write acronyms the way you say them. "S Q L" and "sequel" come out differently, and only you know which your audience uses.
• Kokoro's authors note that voices can sound weak on very short lines and rush on very long passages. Avoid one-word clips and break a wall of text into paragraphs.
• On Orpheus, Canopy Labs documents emotive tags you type into the text: <laugh>, <chuckle>, <sigh>, <cough>, <sniffle>, <groan>, <yawn> and <gasp>. Test one on a short line first.

No script yet? Write it in AI4Chat chat and ask for "short spoken sentences, numbers written as words". Fixing wording in text is cheaper than regenerating audio.

Not with dedicated voices, not yet. All 20 voices in the tool are English voices, and we'd rather say so here than let you find out after a paid clip.

The one model built for more than English is ElevenLabs. Its documentation lists Hindi and Tamil among the 32 languages its v2.5 models cover; Telugu isn't on that list. Our four ElevenLabs voices are labelled English and we haven't published language-specific voice lists, so if you try Hindi or Tamil text, run one short sentence first and listen to the pronunciation before a long script goes through at the top rate.

If Indian-language narration is most of what you need, a tool with native voices for that language will serve you better today.

It makes sense if you already use AI4Chat for chat or images and want voice in the same account, or if you like to compare engines: one script through Kokoro, Orpheus and ElevenLabs, side by side in one gallery. Pair it with the AI music generator and you can make a narration and a backing track on the same site.

It's the wrong pick for free, private English speech for a one-off: browser-based tools that run Kokoro on your own device need no account. And it's the wrong pick for non-English voices, for the reasons above.