LucadenFree online toolsBrowse toolsPrivate on-device speech AI
Generate an original synthetic English voiceover with selectable voices and adjustable pacing on this device.
Loading tool workspace…
How it works
Important
Keep an untouched copy of important source files. Browser processing depends on available device memory, and you should inspect every downloaded result before deleting the original.
Practical details
Generate an original synthetic English voiceover with selectable voices and adjustable pacing on this device.
Lucaden uses the 82-million-parameter Kokoro text-to-speech model through a browser-only kokoro-js and Transformers.js runtime. Inference runs with WebAssembly and produces 24 kHz mono PCM audio.
The runtime and model revision are pinned and the runtime bundle is verified with SHA-256 before execution. Long input is divided into short segments and joined with small pauses so no sentence silently disappears at the model token limit.
Questions
No. Lucaden downloads the Kokoro browser runtime, quantized model, and selected voice files, then performs inference on this device. The typed text is not placed in those download URLs or uploaded by the tool.
No. Listen for pronunciation, names, numbers, pacing, and artifacts. Clearly disclose synthetic speech where appropriate, and never use it to impersonate a person or mislead listeners.
The pinned browser runtime is about 2.1 MB, the quantized English speech model is about 92 MB, and each selected voice is about 0.5 MB. Compatible browser caches can reuse downloaded assets on later visits.
Open-source foundations
Lucaden documents its standards sources, architectural references, and permissive runtime dependencies while keeping this workspace consistent with the Lucaden interface.