A product that sounds the same in every market is a rare thing, but Qwen3 TTS, now available on Together AI, aims to make it standard. It’s an open text-to-speech model that separates who is speaking (the voice) from how they speak (the delivery), letting you keep a consistent voice across your product—even as you change languages or tone on the fly. Together AI’s serving stack delivers first audio in as little as 97ms, fast enough for real-time agents.
You choose from nine preset voices, each capable of speaking ten languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian. You control the delivery with a plain-language instruction—"excited and upbeat," "calm and reassuring"—and the model adjusts tone and emotion accordingly. Pricing is $20 per 1M characters, with both serverless and dedicated endpoints available.
Control tone and language with a single API call
The model’s core trick is decoupling voice from delivery. You specify the voice, the language, and an instructions field to tell it how to say the line. The same preset voice can deliver the same text as good news or bad, in any of the ten supported languages. Here’s a minimal Python example using the Together SDK:
from together import Together
client = Together()
response = client.audio.speech.create(
model="Qwen/Qwen3-TTS-12Hz-1.7B",
input="Well, that changes everything.",
voice="aiden",
language="en",
extra_body={"instructions": "excited and upbeat"},
response_format="mp3",
)
response.write_to_file("excited.mp3")
If the delivery doesn’t change, double-check that you’ve spelled the instructions field correctly—the API skips unrecognised fields without an error.




