Views
No views yet
onnxruntime-web (WebGPU, with a
WASM fallback).<variant> is pico or nano (see below); the Kanade files are shared.| File | Stage | Notes |
|---|---|---|
<variant>/lm/model.onnx (+ meta.json) | Plapre language model | SmolLM2/LLaMA, autoregressive |
kanade_decoder.onnx (+ .onnx.data) | Kanade decoder | audio tokens + speaker emb → mel |
hift_vocoder.onnx (+ .onnx.data) | HiFT vocoder | mel → 24 kHz waveform |
clone_encoder.onnx | Kanade clone encoder | reference audio → 128-dim speaker embedding |
<variant>/tokenizer.json, config.json | tokenizer / config | BPE + special audio tokens |
<variant>/speakers.json, speaker_proj.json | built-in voices | precomputed embeddings (tor, ida, liv, ask, kaj) |
pico/ and
Nano (hidden 960) under nano/ (e.g. pico/lm/model.onnx,
nano/lm/model.onnx). The Kanade kanade_decoder.*, hift_vocoder.*, and
clone_encoder.onnx are shared across variants and stay at the root. Select a
variant with loadPlapreEngine({ model: "pico" | "nano" }).1import { loadPlapreEngine } from "plapre-in-a-browser";
2
3const engine = await loadPlapreEngine({
4 modelsBaseUrl: "https://huggingface.co/elgehelge/plapre-onnx-web/resolve/main",
5});
6const { samples, sampleRate } = await engine.synthesizeToPcm({
7 text: "Hej, hvordan har du det i dag?",
8 voice: "ida",
9});| Artifact | Upstream | License |
|---|---|---|
<variant>/lm/*, tokenizer.json, config.json, speakers.json, speaker_proj.json | Plapre Pico / Nano by syv.ai | CC BY 4.0 |
kanade_decoder.*, clone_encoder.onnx | Kanade (frothywater) | MIT |
hift_vocoder.* | HiFT, from CosyVoice 2 | Apache-2.0 |
| clone-encoder SSL frontend | WavLM (Microsoft) | MIT |