The Voice Bench

Editorial preview: vendor recordings and hands-on verdicts are pending. What counts as evidence.

TTS pricing explained: bytes, credits and finished minutes

A price per million is incomplete until you know what is being counted. Compare the same script and the same finished workload before choosing a plan.

Updated 2026-09-16Planning guide, not a hands-on test

Start with the billing unit

Fish Audio’s API documentation measures TTS input in millions of UTF-8 bytes. ElevenCreative subscriptions use credits; ElevenAPI separately publishes TTS dollar rates per character. Keep the product, model and payment option with each quote. Sources checked 16 September 2026: Fish Audio API billing, ElevenCreative pricing and ElevenAPI pricing. Read the dated pricing breakdowns for each product.

Characters, bytes and words are different counts. Non-ASCII text can use several UTF-8 bytes per character. Speech duration also depends on pacing. Never turn a headline price into a universal cost per minute by assuming every language and voice behaves like an English sample.

Compare two quotes with one workload

  1. Choose a finished target: for example, your next narrated video, not a vendor’s maximum allowance.
  2. Save the exact script and the target language. Record the unit each quote bills.
  3. Record the plan, model, currency, billing period, included usage and overage terms. Separate an introductory discount from the recurring rate.
  4. During the audition, keep first takes and rejected generations. Measure total generation against the final usable export.
  5. Record editing time separately. A subscription saving can disappear when a difficult name needs repeated corrections.
  6. Use the voiceover cost calculator after converting the quote to generated minutes for your own workload.

A fictional example, not a vendor quote

Suppose a plan costs US$10 per month, includes 30 generated minutes and charges US$1 for each additional minute. You need 30 finished minutes, with 25% extra generation for retakes. That means 37.5 generated minutes and US$17.50 in subscription and usage charges.

If editing takes two minutes per finished minute and your time is valued at US$30 per hour, add US$30. Total: US$47.50 a month, or about US$1.58 per finished minute. These are arithmetic examples, not measured vendor performance or typical retake rates.

Questions to resolve before paying

  • Does a failed or rejected generation consume usage? Are any regenerations free?
  • Does unused allowance roll over, reset, or expire?
  • Is the model you auditioned included in the quote?
  • Can you export the required format on that plan?
  • Are API usage and the web editor billed separately?
  • Does the applicable agreement permit the intended use of that output and voice?

Save the quote and answers with the project. We have not verified an account checkout for either vendor. The Fish Audio vs ElevenLabs comparison keeps published facts separate from untested performance.