A clear price. A voice of your own.

None of those: normalize line endings and Unicode NFC, remove leading and trailing whitespace, then count Unicode code points. Internal spaces, punctuation and newlines count. A Chinese character is not charged by its UTF-8 byte size.

A visible emoji can contain multiple code points. Use “Paste your text” in the calculator for exact local counting, or POST /v1/usage/estimate for an authenticated estimate including available trial characters.

Supported synthesis uses the same $8 per million billable characters. This applies to built-in voices and authorized private clones, in the console and API. There is no monthly API subscription.

Changing the text, voice or format and generating again creates a separate request. Listening to saved previews or downloading a retained result is not a new generation. Upload and private-voice quotas still apply.

Queue waiting is free. Rejected or failed generation is not charged. Funds or trial characters can be reserved while work runs; the reservation is released once execution is confirmed stopped, not simply when a browser disconnects.

For one logical request, keep the same Idempotency-Key and identical input. A retained successful result is reused without another charge. A new key is a new request; reusing an old key for changed input is not a free edit.

A technically successful generation is metered, even if you want a different performance. A quality preference is not the same as a failed API request. Changing voice or text and generating again is a new request at the normal rate.

Start with a short passage before a larger batch. If you suspect a service defect, keep the request ID and usage receipt and contact support. This page does not promise automatic refunds or free re-generation for every unwanted result.

Billing is per input character, not per audio minute. Speaking rate, language, pauses and the text change duration, so a fixed price per minute would be misleading.

Generate a representative short passage and record both its normalized character count and duration. Use that observed ratio for a labeled budget estimate, and add the actual text of each language version. The calculator shows paid character cost before trial credits.

This public calculator estimates the full character price before trial credits. It does not know your wallet, available trial allowance or existing reservations and does not initiate a payment.

Authenticated estimates account for current availability, which can change before execution. The completed request’s usage receipt records billable_characters, trial_characters and charged_usd. Use that receipt to reconcile actual usage.

The current enrollment configuration includes 3,000 trial characters with a 30-day expiry. Daily allowance and voice-registration limits also apply. The console shows your actual remaining allowance and expiry.

Trial characters are used first when valid and available; the remainder uses prepaid API balance. Public saved previews and the homepage demo are separate experiences, not an unlimited API allowance.

Top-ups are prepaid, not a recurring subscription. This release has no automatic recharge and no scheduled expiry for paid credits. Available amounts and the final payment total are shown in Checkout before you pay.

Credits can be spent only on Voice API usage. For payment or unused-credit questions, contact account support with the relevant receipt; do not send your full card details or API key.

No. The reading apps and Voice API share a CastReader identity, but API access, wallet, private voices and usage are managed separately. A reading subscription is not an API entitlement.

Verify your CastReader identity and activate the existing trial without manual approval. Each account receives one trial grant; validity, daily and abuse limits still apply. Then create a server-side API key. Public sample listening needs no activation.

No: a million characters is a pricing unit across requests. Current queued tasks accept at most 500 characters, with 1 concurrent generation per workspace. Longer scripts need application-side splitting, queueing and audio assembly.

This release suits audio you can prepare in advance. Streaming, realtime conversation and latency/uptime SLAs are not included. Check limits before planning a high-volume or latency-sensitive integration.