Speech & VoiceAvailable

Voice Clone API

Zero-shot neural voice cloning and multilingual speech synthesis. Give the model a short reference recording and generate natural speech in that speaker's voice across 11 Indian languages — no per-user model training required.

AI Coding Agent Prompt

Integrate Voice Clone API with AI

Download or copy the prompt tetrax-voice-clone-ai-prompt.txt for @tetrax/sdk and give it to ChatGPT, Claude, Cursor, or Antigravity to auto-generate the integration code in your app.

Voice Clone API Overview

Zero-shot neural voice cloning and multilingual speech synthesis. Give the model a short reference recording and generate natural speech in that speaker's voice — no per-user model training required.

Base URL

Base URL
https://api.tetrax.in/v1
LIVE

How it works

  • Reference audio → voice_id. Upload 10–30 seconds of clean, single-speaker audio to POST /v1/voices/create and receive a reusable voice_id. The clip is transcribed automatically (ASR).
  • Text + voice_id → cloned speech. Pass any text to POST /v1/voices/clone with the saved voice_id — or skip the saved voice and send an inline ref_audio clip.
  • Cross-lingual. The reference language need not match the output language_code — clone an English voice and synthesize Hindi, Tamil, Marathi and more.
  • Streaming & real-time. Use /voices/clone/stream for chunked binary audio, or the /voices/clone/ws WebSocket for low-latency interactive synthesis.
  • Metered billing. Speech generation is charged per 1,000 characters from your prepaid wallet. Creating a voice is free.
  • Consent enforced. Only clone voices you own or are authorised to use.

Authentication

Send your API key in the api-subscription-key header (recommended). A TETRAX x-api-key or a dashboard Authorization: Bearer token are also accepted. The WebSocket channel additionally accepts the api-subscription-key.<key> subprotocol.

curl "https://api.tetrax.in/v1/voices" \
  -H "api-subscription-key: trx_your_key"

Limits, quotas & billing

  • Reference upload (create): max 50 MB; 10–30 s, single speaker recommended.
  • Inline ref_audio on generate/stream: max 10 MB and 15 seconds.
  • Text per request: max 1000 characters — split longer input and synthesize in parts.
  • Voice count is capped by your subscription tier; the cap returns 403 TIER_LIMIT_REACHED. Deleting a voice frees a slot.
  • Output sample rate defaults to 24000 Hz; 8000 / 16000 / 22050 / 24000 / 32000 / 44100 / 48000 Hz are supported.
  • Speech is billed at ₹0.08 per 1,000 characters against your wallet.

Supported languages

The cloned voice can speak any of these 12 languages regardless of the reference clip's language.

as-IN · Assamesebn-IN · Bengalien-IN · English (India)gu-IN · Gujaratihi-IN · Hindikn-IN · Kannadaml-IN · Malayalammr-IN · Marathiod-IN · Odiapa-IN · Punjabita-IN · Tamilte-IN · Telugu

Error envelope

Errors return a structured object with a human-readable message, a machine-readable code and a request_id for support.

Response

200 OK
{
  "error": {
    "request_id": "req_9f2c1a4b",
    "message": "ref_audio must be at most 10 MB.",
    "code": "invalid_request_error"
  }
}

Endpoints at a glance

7 Endpoints
POST
/v1/voices/create

Create a cloned voice from a reference clip (one shot)

api-subscription-key
POST
/v1/voices/clone

Generate speech in a cloned voice (saved voice_id or inline reference)

api-subscription-key
POST
/v1/voices/clone/stream

Streamed synthesis — chunked binary audio

api-subscription-key
GET
/v1/voices

List cloned voices in your workspace

api-subscription-key
GET
/v1/voices/{voice_id}

Get one cloned voice with its transcript and audio URLs

api-subscription-key
DELETE
/v1/voices/delete/{voice_id}

Delete a cloned voice (204 No Content)

api-subscription-key
GET
/v1/voices/clone/ws

Real-time WebSocket synthesis channel

api-subscription-key