54 lines (42 loc) · 2.25 KB

Changelog

All notable changes to com.ponyudev.sherpa-onnx will be documented in this file.

[0.1.0] - 2026-02-21

First public release with TTS, ASR, and VAD support.

Added

Text-to-Speech (TTS)

6 model architectures: Vits (Piper), Matcha, Kokoro, Kitten, ZipVoice, Pocket (voice cloning)
Editor UI for model import with auto-detection and auto-configuration
Int8 quantization toggle
Flexible deployment: Local (StreamingAssets), Remote (runtime download), LocalZip
Matcha vocoder selector with independent download
Cache pooling for audio buffers, AudioClips, and AudioSources

Speech Recognition (ASR)

15 offline architectures: Zipformer, Paraformer, Whisper, SenseVoice, Moonshine, and more
5 online (streaming) architectures with real-time microphone capture
Partial and final result callbacks
Engine pool for concurrent offline recognizer instances
Endpoint detection with configurable silence rules

Voice Activity Detection (VAD)

2 model architectures: Silero VAD, TEN-VAD
Configurable parameters: threshold, min silence/speech duration, window size
VAD + ASR pipeline for segment-based recognition

Library Installer

One-click native library install for Windows, macOS, Linux, Android, and iOS
Update All to re-install every platform at once when changing version
Desktop libraries from NuGet, mobile from sherpa-onnx GitHub releases
Patched iOS managed DLL with DllImport("__Internal") from this repo's releases

Model Importer

One-click import from URL (tar.gz, tar.bz2, zip)
Auto-detection of model type and architecture
Auto-configuration of model paths and profile settings

Platform Fixes

Android: native AudioRecord fallback when Unity Microphone returns silence
Android: StreamingAssets extraction to persistentDataPath with version tracking
Android: locale guard (LC_NUMERIC = "C") to prevent float parsing crashes
iOS: xcframework architecture filtering (device vs simulator)
Unity: silent AudioSource workaround to force microphone recording
Unity: Microphone.GetPosition() polling with configurable timeout
All: built-in sample rate resampler (any rate → 16 kHz)
Android/iOS: async microphone permission request via UniTask

[0.0.1] - Initial

Initial package skeleton.