kizuna-ai-lab/sokuji v0.15.17

on GitHub

What's Changed

Features

• Piper-Plus TTS engine: Add piper-plus as a new local TTS engine, running entirely in-browser via ONNX Runtime Web WASM
• Japanese phonemization: Integrate OpenJTalk WASM for accurate Japanese kanji reading, pitch accent, and prosody
• Multilingual VITS model: CSS10-JA model supporting 6 languages (ja, en, zh, es, fr, pt), ~145MB total download from HuggingFace
• TTS language routing: Wire target language through the TTS pipeline, enabling language-aware synthesis for multilingual models

Improvements

• Worker naming clarity: Rename tts.worker.js / asr.worker.js to sherpa-onnx-tts.worker.js / sherpa-onnx-asr.worker.js to distinguish engine runtimes
• ORT UMD build: Add ort.wasm.min.js to copy script for classic worker compatibility

Fixes

• Fix OpenJTalk ES module import.meta.url patching for classic worker context
• Fix missing lid and prosody_features tensors causing ONNX inference failures
• Fix phonemization output to match the multilingual demo format
• Add piper-plus WASM assets to extension build via viteStaticCopy

Full Changelog: v0.15.16...v0.15.17