CastReader Voice APIAdd speech timestamps
Opt in to v3 English word or measured word-group alignment for MP3, WAV and queued jobs. Download metadata with the audio receipt; no extra alignment fee.
Use timestamps →CASTREADER VOICE API
Private voices. One simple API.
Turn short English text into speech in your own voice. Create a voice in the console, then generate MP3 or WAV from your application.
Early access is available by invitation. Check service status for availability; the limits below describe the API contract, not live capacity.
From reference audio to your application
Clone your voice
Record yourself, invite a friend, or upload an authorized recording. Once the voice is ready, listen to its preview and reuse its ID. Your voices stay private to your workspace.
Create a voice →Generate speech
Send text and a voice ID to one endpoint. Save the complete MP3 or WAV response, with a receipt for the characters and credits used.
Generate audio →Build with an API key
Name and copy a server key, then call the API from any language. Expiration is optional; revoke or replace a key whenever needed.
Manage API keys →Track every request
Estimate cost, see trial and paid usage, download retained audio and retry safely with the same idempotency key.
Understand usage and billing →Model and limits
| Capability | Limit |
|---|---|
| Model | clone-v1 |
| Languages | English (en) |
| Audio formats | MP3 and WAV · 24 kHz mono |
| Audio timestamps | Optional v3 word / measured word groups; seconds; English |
| Text per request | 120 normalized Unicode characters |
| Concurrent generation | 1 per workspace |
| Private voices | 2 active voices; 5 registrations per UTC day |
| Reference upload | Up to 4 MiB; optional transcript up to 600 characters |
| Generated audio retention | 24 hours after completion |
| API rate limit | 60 requests per minute per key, including polling |
| Usage price | $12 per million normalized Unicode characters |
Send model: "clone-v1" and language: "en". MP3 supports speed 0.5–2; WAV uses speed 1. Characters count after line-ending normalization, Unicode NFC normalization and trimming. Each receipt pins the model, voice and price versions.
GET /v1/models returns configured model limits for your integration. Workspace budgets, permissions and current capacity also apply. Failed and deleted voice registrations still count toward that UTC day’s registration limit.
Choose where to start
In your browser
Create a voice in My voices, try it in API playground and check its receipt in Usage.
Open the playground →In your application
Use Bearer authentication with the HTTPS API. Start with cURL, then use any server-side HTTP client.
Copy your first request →Node.js and Python source clients are available for integrations. Download OpenAPI 3.1 for the complete HTTP schemas.
What this release includes
The public product is private voice cloning and complete MP3/WAV text-to-speech, with queued short requests when immediate generation is busy. Streaming, long jobs, realtime sessions and automatic recharge are not open in this release. Their advanced references are separate from the quickstart.
Speech recognition, voice conversion, music, public voice sharing, emotion tags and multi-speaker projects are not supported. Use a recording you own or have explicit speaker permission to clone. Public demo audio does not grant commercial voice rights.
Data and voice permissions · Pricing · Errors and safe retries