How much does AI text-to-speech cost per script?

Up, per started block of 200 characters. A 210-character line is billed as two blocks — $0.02 — the same as a 400-character one. For anything longer than a sentence this rounding is noise, but it does mean that splitting a script into many tiny calls is more expensive than sending it in one piece.

Characters are counted on the text you send, including spaces and punctuation. Style directions written into the text count too, which is a fair trade because they are what makes the delivery sound intentional rather than read aloud.

Roughly 30 spoken words per block, so a cent buys about twelve seconds of speech. A typical 30-second social voice-over is $0.02–$0.03; a two-minute explainer narration is around $0.10.

The output arrives in the same response — there is no job to poll, unlike images and video. That makes speech the easiest thing to script: send text, receive a WAV.

Dialogue is one call, not two. Assign two speakers with their own voices and the model performs the exchange with the timing between the lines intact, which is what makes it sound like a conversation rather than two takes edited together.

It is the cheap half of a UGC-style ad. Veo 3.1 with native audio roughly doubles the price of a clip; a silent clip plus a generated voice-over is usually cheaper and gives you a script you control word for word. Gemini Omni Flash is the exception — its audio is always included in the per-second rate and lip-syncs, so for talking heads it is worth generating in-model.

Voice-over is also the part you will iterate most. At a cent per 200 characters, ten takes of a 30-second read cost less than a single Veo generation.