Qwen3-TTS Base (clone) API

Alibaba Cloud, open modelqwen3-tts-baseModel source: Hugging Face

Speaks in the voice you want: give it a few seconds of sample audio and it imitates it.

Qwen3-TTS Base turns text into speech in a voice cloned from a sample you provide: between 1 and 30 seconds of audio plus its literal transcript. It costs €5.00 per million characters, on the same /v1/audio/speech call as the OpenAI API.

What it does

Text to speech

tts

How it is charged

Each million characters

5.00 €

Where it runs

Spain (EU)

MIG B200

Brain size

Small and fast

TTS, Voice cloning

What it is good at

A brand voice that is yours. Clone a voice from a short sample and use it everywhere: assistant, IVR, videos, audiobooks. Consistent across content and time.
Natural prosody. Pauses, emphasis and intonation that sound spoken, not read.
Spanish and nine more languages. Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian.

You are responsible for having the rights to any voice you clone. It has no ready-made voices: for one, use Qwen3-TTS CustomVoice.

Pricing

€ per million characters, VAT included

Each million characters (use)5.00 €

try it with 5 € freeActivated with your card, and you only pay for what you use.

For your technical team

You send the text and get the audio back. You pick the model and the voice.

curl https://api.gpuflow.ai/v1/audio/speech \
  -H "Authorization: Bearer $GPUFLOW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "qwen3-tts-base",
        "input": "Hola",
        "ref_audio": "https://example.com/muestra.wav",
        "ref_text": "transcripción literal de la muestra" }' --output speech.wav

Limits and formats

Requests per minute: 1,000 per key, with bursts up to double; raisable to 5,000 from the dashboard.
Output formats: wav (default), mp3, flac, pcm, aac and opus.
Cloning sample: 1 to 30 seconds of audio, with its literal transcript in ref_text.

Qwen3-TTS Base (clone) compared with similar models

	Qwen3-TTS Base (clone)	Qwen3-TTS CustomVoice
Best for	Your own voice from a short sample	Ready-made voices with adjustable tone and emotion
Voices	Clones your sample	Nine ready-made voices
Price	5.00 € per million characters, VAT included	19.00 € per million characters, VAT included

A ready-made voice with control over tone: Qwen3-TTS CustomVoice; your own voice from a sample: Qwen3-TTS Base. Both use the same endpoint.

Frequently asked questions

How much does Qwen3-TTS Base (clone) cost?

€5.00 per million characters, VAT included. No monthly fee.

Where does it run?

On our NVIDIA B200 GPUs, in a data centre in Spain.

How do I clone a voice?

Call /v1/audio/speech with model qwen3-tts-base, the text in input, the sample in ref_audio (an https URL or a data:audio/wav;base64 URI) and its literal transcript in ref_text.

What sample do I need?

Between 1 and 30 seconds of clear speech; outside that range it is rejected. WAV is the recommended format.

Can I clone without the transcript?

Yes, with x_vector_only_mode: true, but it only copies the timbre and the quality is lower. With ref_text the result is better.

Is the sample charged or stored?

You pay only for the characters of input; ref_text is not charged. The sample is processed in memory for that call and we do not keep it.