How does OpenAI charge for TTS?

OpenAI bills text-to-speech per character of input text, not per minute of audio. tts-1 costs less per character and tts-1-hd costs more in exchange for higher audio fidelity. Spaces and punctuation count toward characters.

How long will my audio be?

Spoken English runs around 150 words per minute at a normal pace, so this tool estimates duration from your word count. Actual length varies with the voice, the speed parameter, and how many pauses your text contains.

What is the difference between tts-1 and tts-1-hd?

They share the same voices, but tts-1-hd renders cleaner, higher-quality audio at higher latency and cost. Use tts-1 for real-time or high-volume jobs and tts-1-hd for polished, published audio.

How many characters is a word?

English averages roughly 6 characters per word including the trailing space, so a 1,000-word script is about 6,000 characters. This calculator uses that ratio when you enter words.

Are these prices exact?

They reflect commonly published OpenAI per-character speech pricing, but rates change. Always confirm the current numbers on the OpenAI pricing page before budgeting a large job.

OpenAI TTS Cost Calculator

OpenAI TTS cost calculator

OpenAI’s speech API charges per character of input text, so the cost of an audiobook or a podcast intro comes down to script length and which model you pick. This calculator converts your word or character count into a USD estimate for both tts-1 and tts-1-hd, plus the expected audio duration.

How it works

The math is straightforward:

characters = words × 6                  (≈ chars per English word)
cost       = characters × price_per_char × quantity
duration   = words / 150 wpm            (normal speaking pace)

tts-1 is the budget, lower-latency model; tts-1-hd costs more per character for higher fidelity. The calculator shows both so you can see exactly what the quality upgrade costs for your volume.

Tips for controlling TTS spend

Edit before you synthesize. You pay for every character of input — tighten the script first.
Reserve HD for published audio. Use tts-1 for drafts, previews, and real-time agents; switch to tts-1-hd only for the final render.
Reuse static segments. Generate fixed intros and disclaimers once and splice them in rather than re-synthesizing each run.
Batch and cache. Identical inputs produce billable calls every time — cache the output audio, not just the text.