---
title: "Qwen3-TTS | Tarka Utilities"
description: "Apache 2.0 text to speech with consent-backed voice cloning."
url: "https://tarkahq.com/utilities/qwen3-tts/"
---

Text to speech and voice cloning

# Qwen3-TTS

Apache 2.0 text to speech with consent-backed voice cloning.

[Sign up](https://tarkahq.com/console?auth=signup) [Utility model docs](https://tarkahq.com/docs/inference/utility-models/)

- **Status**: Live
- **Languages**: 10 languages
- **License**: Apache 2.0

## Model ID

| Model ID | Designed for |
| --- | --- |
| `qwen3-tts` | Text to speech with consent-backed voice cloning |

Confirm access with `GET /v1/models`, then follow the [utility-model docs](https://tarkahq.com/docs/inference/utility-models/) for the canonical v1 REST or native gRPC request contract.

## Pricing

`qwen3-tts`: Free for a limited time. ~~$0.0004~~ $0.00 per audio second

 USD · catalog 2026-11-04.1. Release verified 2026-11-04. EUR uses the catalog conversion. [Live catalog](https://tarka.rest/control/v1/pricing).

## Overview

Tarka’s `qwen3-tts` route uses `Qwen/Qwen3-TTS-12Hz-1.7B-Base` for voice cloning. Create a clone from a clean recording of a consenting speaker, then pass the returned `voice_…` ID to the same `qwen3-tts` model at `/audio/speech`.

This is the Qwen3-TTS Base model, not Qwen-Audio-3.1. It requires a saved clone ID; built-in speaker names and free-form voice-design instructions belong to other model variants. The current route returns completed audio, not a realtime voice session or streaming audio.

The upstream model documents Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. Nepali is not in that list; use `nepali-parler-tts` for the Nepali speech route and assess output quality on your own text.

## Voice cloning

Create clones through canonical v1 REST or the matching native gRPC methods. Every create request requires a consent name and date. Use a clean reference recording no longer than 30 seconds. The model can use a short sample, but the quality of a particular clone is not guaranteed by its duration.

Follow the [create, read and delete examples](https://tarkahq.com/docs/inference/utility-models/#clone-title) to obtain your own ready clone ID before running the speech request below.

```
curl --fail-with-body https://tarka.rest/v1/audio/speech \
  -H "Authorization: Bearer $TARKA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "qwen3-tts", "voice": "voice_0123456789abcdef0123456789abcdef", "input": "Hello world.", "response_format": "mp3" }' \
  > out.mp3
```

Clone IDs are model-specific and private to the organization that created them. Delete a clone when it is no longer needed. Use model discovery and your organization limits as the source of truth for access and request limits. Supported audio output formats are `wav`, `mp3`, `flac`, `opus`, `aac` and raw `pcm`.

Changing the serving machine does not require recreating a saved clone. Keep using its existing ID with the same model and organization. See [service status](https://status.tarkahq.com/) for current health.