Gradium TTS public beta: open for your hardest cases

A new Gradium TTS model is available today in public beta. We continue to improve versus the current production model, handling natively the complex cases we prioritized for our customers, with no pre-processing or text normalization on your side. Anyone can test it now, and we are offering 1M Gradium credits to everyone who sends us feedback during the beta.

What the beta model improves

In our June release, we focused on pronunciation accuracy for the cases that break voice agents in production. This beta continues that work, handling the complex cases customer support agents run into:

Phone numbers. National and international formats, grouped the way a native speaker would say them.
Email addresses. Character-accurate, including dots, hyphens, and domains.
IBAN numbers. Read digit by digit with correct grouping, in every supported language.
Time expressions. "halb drei" (2:30), "viertel vor acht", "14:30 Uhr" pronounced natively.
And many more. Account handles, measurements, reference codes, license plates, URLs, ordinals. Where the model doesn't handle a case yet, that is exactly what this beta is for.

Hear it in practice

Phone numbers

Reach our London office on +44 20 7946 0958, our Paris desk on 0033 1 42 86 82 00, or dial 311 locally for enquiries.

Gradium TTS beta

Reference codes and spelling

Reference RF24, surname spelled F-O-R-S-Y-T-H.

Gradium TTS beta

The goal is to make life as easy as possible for developers building agents: send the text as it comes out of your LLM or your database, and the model reads it correctly. No prompt engineering, no normalization rules to write and maintain.

This is still a beta. We are building it with the developer community: the cases you hit in production define what the model learns to handle next.

How to access the beta

The beta model is available through the API only. Set the model name in your setup message. With the Python SDK:

python

import gradium

client = gradium.client.GradiumClient(api_key="your-api-key")

result = await client.tts(
    setup={
        "model_name": "gradium-tts-beta",
        "voice_id": "Bla6SbVMczYnOhfK",
        "output_format": "wav",
    },
    text="This is Gradium.",
)

with open("output.wav", "wb") as f:
    f.write(result.raw_data)

The same model_name works over the WebSocket API for real-time streaming. Your existing voices, including custom voices, work as-is. The current production model remains the default; nothing changes unless you opt in.

If you use a coding agent, paste this prompt to wire up the beta model in minutes (you need a Gradium API key first):

prompt

Install the Gradium Python SDK and write a script that synthesizes speech
with the Gradium TTS beta model (model_name: "gradium-tts-beta",
voice_id: "Bla6SbVMczYnOhfK", WebSocket endpoint
wss://api.gradium.ai/api/speech/tts, API key from the GRADIUM_API_KEY env
var). Use the docs at
https://docs.gradium.ai/api-reference/endpoint/tts-websocket#setup.

1M Gradium credits for feedback

Every participant who provides feedback during the beta receives 1M Gradium credits.

Test the beta model through the API as shown above.
Send us your findings in the support channel of our Discord, or through the contact form: the input text, the language, the voice ID or name, and what you expected versus what you heard. We are looking for feedback grounded in your real production use cases; failure cases are the most valuable.
Credits are applied to your account once your feedback has been processed.

Try the beta, and send us your hardest cases on Discord to claim your 1M credits.