AI Voice Designer (Beta) — Design a Voice from a Text Description

How Voice Designer works — the four-step flow

The design flow is intentionally simple. You do not need any audio equipment, any knowledge of voice synthesis parameters, or any prior experience with TTS tools. The entire process — from description to saved, usable voice — takes about two minutes and is self-service inside your account.

1. 1

   Describe the voice you want

   Open the Design a voice tab in your account. Type a free-text description: for example, “warm, female, British accent, slightly slower than normal pace” or “energetic young male, American, fast delivery, slightly higher pitch”. You can be as specific or as vague as you like — the system interprets natural language, not structured parameters. There is no required format or vocabulary.

2. 2

   Receive three candidate previews

   The system maps your description to three diverse candidate recipes — each combining a base voice from EasyVoice's catalogue, a speed adjustment, and a pitch-shift — and synthesises a short audio sample for each candidate on a fixed test sentence. The three candidates are intentionally distinct: if you described a calm deep male voice, you might get one candidate that leans deep-and-slow, one that is deep-and-neutral, and one that adds a slight pitch-down while keeping speed natural. This spread gives you a genuine choice rather than three near-identical results. The previews are real synthesis output, not mock clips — what you hear is exactly what you will get when you synthesise your own text.

3. 3

   Pick your favourite, name it, and save

   Listen to the three candidates, pick the one that best matches your intent, give it a display name (e.g. “Podcast Host” or “Product Demo Voice”), and click Save. The voice is stored to your account with a stable ID prefixed vd_. Pro accounts can save up to 10 designed voices at a time; you can delete and replace voices if you need to experiment further.

4. 4

   Use it everywhere in EasyVoice

   Saved designed voices appear in the voice selector under a “Designed Voices” group alongside standard voices. They are available in the TTS text editor, in podcast dialogue mode (assign to any speaker role), and via the REST API — pass the vd_ voice ID to the /v1/audio/speech endpoint exactly as you would any other voice ID. The response format, streaming support, and request shape are identical to standard API calls.

Honest Beta framing — what “Voice Designer” is and is not

We call this feature Voice Designer (Beta) and we want to be explicit about what that means. The current implementation does not generate a novel voice from scratch in the way that a large-scale generative audio model would. Instead, it maps your text description to a structured recipe — a combination of one of EasyVoice's 56 catalogue voices, a speed multiplier, and a pitch-shift applied in post-processing — and synthesises a preview using that recipe. The mapping logic uses Claude to interpret the description and select three diverse, non-overlapping candidates from the space of possible combinations; if the API key is unavailable, a keyword-based fallback produces reasonable candidates without any external call.

This matters because the label “AI Voice Designer” could reasonably imply a fully generative model. We are choosing to be transparent rather than use language that overstates the capability. What you actually get is useful in practice: the description-to-preview flow saves you the time of manually trying voices and tweaking parameters, the three-candidate spread gives you a genuine choice, and the saved voice is stable and portable. But it is a guided search through a structured parameter space, not the synthesis of a unique voice that has never existed before.

The “(Beta)” qualifier signals that we plan to replace the recipe system with a genuinely generative model — one that can produce voices outside the existing catalogue — once the hardware constraints are resolved (current constraint: generative voice models do not fit in the 8 GB VRAM available on the inference server). When that upgrade ships, designed voices created under the Beta recipe system will be migrated or flagged clearly. In the meantime, the feature is useful, the copy is honest, and the pricing reflects it: Voice Designer Beta is included in the standard Pro plan at $9.99/mo, not gated behind a premium tier.

Who uses Voice Designer and for what

The most immediate use case is branded content narration. A marketing team, podcast producer, or e-learning author has a specific vocal identity in mind — a tone of voice that fits their brand — but cannot easily express it by browsing a catalogue. Voice Designer lets them describe the target and iterate quickly: try a description, listen to the candidates, refine the description, try again. Within a few iterations you typically converge on a voice that is close enough to ship; the saved voice then becomes a consistent asset used across all content for that project.

Podcast dialogue mode is a particularly natural fit. When producing a scripted multi-speaker podcast, you want the two speakers to sound clearly distinct — different gender, accent, or vocal energy — but you also want them to feel like they belong in the same episode. Voice Designer lets you describe both roles in relation to each other: “warm British female, moderate pace” for the host and “younger American male, slightly faster, higher energy” for the guest. You preview both, save both, assign them to Speaker A and Speaker B in the dialogue editor, and generate the full episode without recording a single audio sample.

API integrations that need a voice selection step in their workflow benefit from Voice Designer because the design flow is fully REST-API accessible. If you are building a SaaS product that generates narrated content for end users and you want to give those users a voice customisation step without exposing the full voice catalogue, you can expose a simple description field in your product UI, call the EasyVoice design endpoint on their behalf, present the three candidates, and save the chosen voice to their account. The savedvd_ voice ID is then used for all subsequent synthesis calls for that user.

Iterative content testing is a lower-stakes but surprisingly common use case: which vocal character performs better in a short social video? Voice Designer lets you create two or three distinct named voices for the same script, generate audio for each, run the variants, and retire the underperformers. Because designed voices have stable IDs, you can keep the winning voice in production and delete the others without disrupting anything.

Voice Designer is on the Pro plan

Voice Designer is intentionally included in the standard Pro plan, not gated behind Pro+ ($19.99/mo). Pro+ is reserved for voice cloning — a separate feature that enrolls a real recorded voice sample with consent attestation and AudioSeal watermarking. If your use case is describing and saving synthetic voices rather than cloning a real voice, Pro is the right tier.

Related features

Frequently asked questions