MeowRVC

KO | EN

gitlite — search

#artificial-intelligence #git #privacy #agent #pdf #animation #pytorch #postgresql #mysql #php

(★ 9)

AI real-time voice changer for Android / 安卓实时AI变声器

파일 탐색기

ZIP 다운로드

- - android.yml
- build.gradle.kts
- - gradle-wrapper.jar
  - gradle-wrapper.properties
- libs.versions.toml
- build.gradle.kts
- proguard-rules.pro
.gitattributes
.gitignore
.gitmodules
ALL_SOURCE.md
build.gradle.kts
CHANGELOG.md
gradle.properties
gradlew
gradlew.bat
jitpack.yml
LICENSE
README.en.md
README.ja.md
README.ko.md
README.md
settings.gradle.kts
THIRD_PARTY_NOTICES.md

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

AI real-time voice conversion app for Android. Runs RVC (Retrieval-based Voice Conversion) models locally using ONNX Runtime.

## Features

- **Real-time voice conversion** - Select a model, speak into the mic, hear the converted voice instantly
- **Floating mic overlay** - Record audio from any app, process through AI, auto-playback through speaker for other apps to capture
- **Auto-detect microphone usage** - When WeChat/QQ starts recording, automatically plays the latest processed audio through the speaker at max volume (acoustic coupling)
- **Adjustable latency** - 0.5~5 seconds, lower for real-time, higher for better quality
- **Pitch shift** - +/- 15 semitones
- **ONNX Runtime acceleration** - Supports NNAPI/NPU/GPU

## How It Works

```
1. Select model -> Start floating mic overlay
2. Tap overlay to record -> Stop -> RVC processes -> saves to /sdcard/rvc/
3. Open WeChat/QQ -> Send voice message
4. App detects mic is busy -> Auto-plays latest processed audio through speaker at max volume
5. WeChat's mic picks up the speaker output -> Friend hears the converted voice
```

## Requirements

- Android 8.0+ (API 26+)
- Root (Magisk) for `setCommunicationDevice` / speaker routing
- ~4GB free RAM for model inference
- RVC model files (`.onnx`) in `/sdcard/models/`

## Build

```bash
git clone https://github.com/xiaoxiaoyu-miao/MeowRVC.git
cd MeowRVC
./gradlew :app:assembleDebug
```

Requires: Android SDK 35, NDK 27, JDK 17

The `onnxruntime.jar` and `.so` files are not included in the repository (`.gitignore`). Download them from Maven Central or GitHub Releases:

- `onnxruntime-android-1.18.0.aar` → extract `classes.jar` to `voicechanger/libs/onnxruntime.jar`
- Extract `jni/arm64-v8a/libonnxruntime*.so` to `voicechanger/src/main/jniLibs/arm64-v8a/`

## Model Files

Place your RVC ONNX model folder under `/sdcard/models/`. The app scans for:

- `hubert.onnx` - ContentVec/HuBERT feature extractor (v2, 768-dim)
- `text_encoder.onnx` - Synthesizer text encoder
- `flow.onnx` - Normalizing flow (reverse)
- `generator.onnx` - HiFi-GAN NSF generator
- `config.json` - Model metadata

Convert your `.pth` RVC model on PC:

```bash
python convert_rvc.py --rvc_root RVC-Project --synthesizer model.pth --output_dir output
```

## Tech Stack

- **ONNX Runtime** - Model inference
- **Android AudioRecord/AudioTrack** - Audio capture and playback
- **AudioManager.AudioRecordingCallback** - Microphone usage detection
- **setCommunicationDevice** - Speaker routing (Android 14+)
- **Material 3** - UI

## License

MIT

MeowRVC

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README