# CastReader Voice API

> A REST text-to-speech API for short English audio with private, authorized cloned voices.

## Current product facts

- Canonical website: https://voice.castreader.com
- Product owner: CastReader, https://castreader.com
- Access: developer preview by invitation; request access on the website.
- Language: en.
- Model alias: clone-v1.
- Formats: complete MP3 and WAV audio files.
- Optional speech timestamps: return_timestamps=true uses qwen-pcm-measured-groups-v3 for English word or measured word-group alignment. Read GET /v1/requests/{request_id}/timestamps. No extra fee; optional alignment may be unavailable while successful audio is charged once.
- Queued speech: enabled, POST /v1/jobs; maximum 120 normalized characters; 600 second task deadline.
- Concurrency: 1 generation per workspace. Availability is shared and variable.
- Price: USD 12 per million billable Unicode code points after NFC normalization. Internal spaces and punctuation count. This is not UTF-8 byte billing.
- Trial: 3000 characters after activation, valid for 30 days; daily allowance applies.
- No monthly API subscription. API wallet and voices are separate from CastReader reading apps and Pro.
- Successful retained result: downloadable for up to 24 hours without another generation charge.
- Not currently offered: real-time speech, streaming, long-form requests, a latency/uptime SLA or an EU-region deployment.
- One website, account, API key and prepaid wallet. Account, billing and resource metadata use the US control plane. The dispatch API selects China for mainland-China networks and the US elsewhere. Text, reference recordings, inference and private audio stay in the selected region, with no cross-region voice-data fallback.

## Primary sources

- [API overview](https://voice.castreader.com/): Product workflow, fit and current limits.
- [Pricing](https://voice.castreader.com/pricing): Character metering, trial, prepaid balance and calculation examples.
- [Developer documentation](https://voice.castreader.com/docs): Authentication, voices, queued tasks, retries and usage.
- [Speech timestamps](https://voice.castreader.com/docs/timestamps): v3 alignment, response examples, retention and billing.
- [Capabilities](https://voice.castreader.com/docs/overview/capabilities): Current enabled capabilities.
- [OpenAPI 3.1](https://voice.castreader.com/openapi.json): Machine-readable request and response contract.
- [Data and voice permissions](https://voice.castreader.com/data): Processing regions and voice authorization.
- [Request access](https://voice.castreader.com/request-access): Developer application.

## Product questions

### What is the CastReader Voice API?

CastReader Voice API is a REST text-to-speech service for generating English audio with private, authorized cloned voices. It returns MP3 or WAV and provides API keys, queued tasks, usage records, and a separate prepaid wallet.

### How is the API different from the CastReader reading apps?

The browser extension and mobile apps help people listen to content. Voice API lets developers generate audio inside their own products. You use your CastReader identity, but API access, credits, voices, and usage are managed separately. A Pro reading plan does not include API usage.

### How much does text to speech cost?

Successful generation costs $12 per million billable characters, with no monthly API subscription. A paid 100-character request costs $0.0012. Available trial characters are used first. Characters are Unicode code points after NFC normalization, including internal spaces and punctuation; this is not UTF-8 byte billing.

### Can I use my own voice or a voice someone shares with me?

You can create a private voice from an authorized recording. The console offers self-recording, a recording invitation, and audio upload. You need the speaker’s permission for cloning and your intended use. Personal voices from the reading apps are not automatically imported into API workspaces.

### Is this a real-time or long-form speech API?

The current preview is for short English speech and queued workflows. It does not offer real-time conversations, streaming, or an uptime or latency SLA. Check the published capabilities for current input limits before integrating.

### What happens if I retry or close the browser?

A queued task persists when you leave the page. Save its job ID and the original Idempotency-Key. Repeating the same input with that key returns the same task; downloading a retained successful result does not generate or charge again. Waiting and tasks cancelled before execution are free.

### Where is API data processed?

One website, account, API key and prepaid wallet. Account, billing and resource metadata use the US control plane. The dispatch API selects China for mainland-China networks and the US elsewhere. Text, reference recordings, inference and private audio stay in the selected region, with no cross-region voice-data fallback. International voice data is separate from the China service, with no automatic cross-region fallback. See the data and voice permissions page for handling and retention rules.