- Published on
Write a Prompt, Get a Brand New Synthetic Voice in Seconds
- Authors
- Name
- DeepHorus
Paris-based voice AI company Gradium, spun out of the Kyutai research lab, has launched Voice Design, a tool that generates entirely new synthetic voices from a written text prompt in seconds. This directly addresses a major limitation for voice agent teams, who often struggle with the limited selections in traditional voice catalogs and the complex rights management involved in voice cloning. The new feature allows developers to bypass these constraints by simply describing the desired voice characteristics, such as age, accent, and energy, to create a unique voice on the fly.
Voice Design is now live and integrated into Gradium's API and Studio interface, available for free across all plan tiers, including the free tier. Developers can submit a description of up to 500 characters in one of five languages (English, French, Spanish, Portuguese, or German) to generate up to five candidate voices. The process is non-deterministic, meaning the same prompt can yield different voice variations. Promising a production-ready workflow, the company reports latency of 3 to 5 seconds for candidate generation.
The technology works through a specific API flow. An initial request generates candidate IDs, which are then polled until ready. Each candidate can be auditioned using the standard text-to-speech endpoint. Finally, a developer can promote a chosen candidate to a permanent, production voice. These kept voices share a custom voice slot with cloned voices and are subject to the plan's limit: one on the free tier and up to 1,000 on paid plans.
To validate its performance, Gradium conducted a vendor-designed blind listening test. Across 7,627 comparisons in five languages, native speakers preferred Gradium's generated voices 72.6% of the time against five competing systems, including ElevenLabs. The system showed its strongest advantage on complex regional accents, achieving a 97% preference rate for Quebecois French. An automated judge using Gemini 3.1 Pro produced the same ranking.
The launch positions Gradium as a strong contender in the competitive voice AI market. By eliminating the need for reference audio and simplifying voice creation, it lowers the barrier for developers needing specific, custom voices for applications like localized customer service or narrative content. The tool is presented as a direct solution to the "casting brief" problem where no catalog voice is a perfect fit.
Industry observers will be watching the adoption of Voice Design and how it performs in real-world deployments beyond benchmarks. The company's claim of strong prompt adherence, backed by its score on the InstructTTSEval academic benchmark, suggests a focus on precise control. This move could accelerate the shift from using pre-made voice catalogs to generating bespoke voices as a standard practice in the audio industry.
