Plapre in a browser

Open-source Danish text-to-speech, running fully in your browser — no server, no cloud, no API key. WebGPU with a WASM fallback.

Compute backend

The model can run on the GPU (WebGPU) or the CPU (WASM).

Detected availability

WebGPU: …

multi-threading: …

LM Decoder+vocoder

Auto = the decoder+vocoder run on WebGPU when available; the LM runs on multi-threaded WASM when possible (it's faster there), otherwise WebGPU.

Model & artifacts

Model Models base URL Where the converted ONNX artifacts are served from. Defaults to /models. Variant-specific files (the LM, tokenizer, and speaker tables) load from the model's sub-path; the shared Kanade decoder/vocoder sit at the root. Point it at a release/CDN bundle to run without a local copy.

Speech synthesis

Voice Speed 1.00× Temperature 0.80

Text (Danish)

Voice cloning

Clone a voice from a short reference clip (10–30s of clean speech). The audio never leaves your browser.

Console