AI Voice Maker 🎛️
Two ways to create: blend — drag across a 2D map to morph between 54 real
Kokoro voices — or adjust — start from one voice and tune traits like depth,
warmth, and brightness along real voice directions. Preview instantly, save to My Voices,
share as a link, or download the .bin style file. Everything runs in your
browser — no signup, no API key, no uploads.
How it works
Explore the Map
Each dot is a real Kokoro voice — hover to see its identity, click to select it exactly.
Drag to Morph
Drag the target into the space between voices — the nearest voices blend by distance.
Preview
Load the model once (~305MB, cached), type a test sentence, and hear your new voice.
Save, Share, Export
Save recipes in your browser, share via link, or download the .bin style file.
How the voice map works
Each Kokoro voice is a .bin file of style embedding vectors. The map lays all 54 voices out with t-SNE so similar-sounding voices cluster together. When you drag, the tool blends the nearest voices by inverse distance over the real embeddings — not reconstructed approximations — so dragging between two close dots yields a mix of two similar voices, and previews always use the proven weighted-average path.
Language routing: the dominant (highest-weight) voice determines the phonemization path. English synthesizes fully locally; other languages send the test text to the documented phonemization endpoint before on-device synthesis. Cross-language blending is off by default — enable it only if you know you want the mixed result.
Use your custom voice for
Experimental Tool Boundaries
The Voice Maker is an experimental embedding mixer, not a voice-cloning, training, or identity-replication service. It calculates distance-weighted combinations of existing style arrays; it does not learn from a microphone sample, infer a real person's vocal identity, or create a new language model. A blend can sound intermediate, muffled, or unexpectedly similar to one dominant source voice.
Download a blend for a matching Kokoro-82M voice-loading workflow — it is not a
universal voice file for other TTS engines. Keep the recipe (name, weights, language)
with the file so you can reproduce or retire it later. The .bin output
contains style values, not generated speech, a recording, or an automatically granted
commercial license.
How to Evaluate a Designed Voice
Use a fixed test passage with short and long sentences, punctuation, names, numbers, and questions in the language you will actually publish. Listen to the source voices first, then the blend at the same model precision, speed, browser, and backend. Change one thing at a time — weights, position, or text — so an improvement can be attributed to a specific adjustment.
Check intelligibility, pronunciation, pacing, loudness changes, high-frequency artifacts, and consistency across several sentences. Do not judge a voice from a single greeting, and review the non-English text handling in the Privacy Policy before testing confidential text.
Output, Rights, and Data Handling
The browser downloads the required model and base-voice assets, computes the blend on the device, previews local speech, and creates the downloadable file without uploading the generated audio. English Kokoro uses its local phonemizer after required assets are available; non-English Kokoro sends entered test text to the documented phonemization endpoint before local synthesis. Model delivery and ordinary site analytics are separate network activity. You are responsible for the upstream terms attached to each source voice, your test text, any person or character the result may evoke, and the context in which generated audio is published.
Frequently asked questions
What is the OfflineTTS Voice Maker?
It is a free browser tool for designing your own synthetic voice. Each Kokoro TTS voice stores its character as a style embedding; the Voice Maker lays those voices out on a 2D map and lets you drag a target position to blend nearby voices into a new timbre. Preview the result instantly and export the style file.
Do I need an account or an API key to create a voice?
No. Voice creation is 100% in your browser: no signup, no login, no API key. Saved recipes live in your browser’s IndexedDB, and share links are just encoded recipes.
Can I make a voice that sounds like a specific person?
The map blends existing Kokoro voices; it does not learn from a recording or replicate a real person’s identity. To clone from a reference sample you have rights to, use the Pocket TTS cloner in the Clone tab — record or upload 3–10 seconds and the engine builds a custom voice on-device.
Is the downloaded .bin file a finished voice?
The .bin file contains the blended style embedding (numbers), not generated speech. It is intended for voice-loading workflows compatible with the Kokoro-82M ONNX model. Blending is an experimental technique — results can be intermediate, muffled, or unstable.
Which languages can a designed voice speak?
The recipe inherits the language of its dominant source voice. English text phonemizes locally; other languages (Japanese, Mandarin, Spanish, French, Hindi, Italian, Brazilian Portuguese, British English) phonemize through the documented OfflineTTS endpoint before local synthesis. Cross-language blends are off by default because they often sound unstable.
Can the Voice Maker clone my voice?
The Design tab blends existing voices. For zero-shot cloning of a reference sample, use the Clone tab — record or upload 3–10 seconds and the Pocket TTS engine builds a custom voice on-device.
Does voice design work offline and stay private?
English previews synthesize fully on-device after the model and base voices download. Non-English text sends only the test text to the documented phonemization endpoint before local synthesis; the generated audio and your saved recipes never leave the browser.
Can I use my designed voice in the main TTS tool?
Yes. Save a recipe here and it appears in the main TTS workspace under the “My Designs” group in the voice picker — up to 8 designs are installed automatically.
Keep exploring
Browse all 54 base voices or open the main TTS workspace to use a saved design.