更新 readme

2026-04-18 11:42:08 +08:00
parent 1ad70525c1
commit 3f84c111d9
4 changed files with 730 additions and 833 deletions
@@ -1,405 +1,404 @@
-# Duix Mobile for Android SDK Documentation
+# 安智伴 · AI 数字人陪护终端（Android）

 English | [中文](./README_zh.md)

-## 1. Product Overview
-
-`Duix Mobile for Android` is a lightweight, fully offline 2D digital human solution for Android, supporting real-time rendering of digital avatars driven by voice audio.
-
-### 1.1 Application Scenarios
-
- **Low deployment cost**: Suitable for unattended scenarios such as large-screen terminals, government halls, and banks.
- **Minimal network dependency**: Runs entirely locally, no internet required, stable operation in subways and remote areas.
- **Diverse functionality**: Can serve as a guide, Q&A customer service, intelligent companion, and more.
-
-### 1.2 Core Features
-
- Customizable digital avatar and local rendering
- Real-time voice-driven playback (supports WAV playback and PCM streaming)
- Motion playback control (specific or random actions)
- Automatic resource download management
+> 本项目与 [digital-human-monitor-v2](https://github.com/user/digital-human-monitor-v2)（家属监护 Web 端）配合使用，共同构成完整的"AI 数字人 + 家庭远程关怀"解决方案。

 ---

-## 2. Terminology
+## 目录

-| Term               | Meaning                                                                    |
-|--------------------|----------------------------------------------------------------------------|
-| PCM                | Pulse-Code Modulation, raw audio stream with 16kHz sample rate, 16-bit depth, Mono channel |
-| WAV                | An audio file format that supports PCM encoding, suitable for short voice playback |
-| RenderSink         | Rendering data reception interface, implemented by the SDK, can be used for custom rendering or default display |
-| DUIX               | Main control object of the digital human, integrates model loading, rendering, broadcasting, and motion control |
-| GLES               | OpenGL ES, a graphics interface for rendering images on Android               |
-| SpecialAction      | A JSON file attached to the model that marks action intervals (e.g., greetings, waving) |
+- [需求分析](#需求分析)
+- [设计思路](#设计思路)
+- [技术架构](#技术架构)
+- [模块说明](#模块说明)
+- [全链路数据流](#全链路数据流)
+- [通信协议](#通信协议)
+- [集成要求](#集成要求)
+- [SDK 接口文档](#sdk-接口文档)
+- [构建与运行](#构建与运行)

 ---

-## 3. SDK Access
+## 需求分析

-### 3.1 Module Reference (Recommended)
+### 背景

-1. Obtain the complete source package, unzip it, and copy the `duix-sdk` directory to the project root directory.
-2. In the project `settings.gradle`, add:
+中国老龄化进程加速，独居或空巢老人的心理陪伴和安全监护成为刚需。子女因工作或地理距离无法陪伴，需要一种低门槛、有温度的技术方案来弥合亲情距离。

-```gradle
-include ':duix-sdk'
+### 核心需求
+
+| 需求维度 | 描述 |
+|---------|------|
+| **AI 陪伴对话** | 老人可随时与屏幕上的"数字人"自然语音对话，获得情感陪伴和生活助理服务 |
+| **家庭留言桥梁** | 子女通过 Web 端发送关怀留言 → 数字人主动转达；老人口述留言 → 子女在 Web 端收到 |
+| **使用行为上报** | 设备自动向云端上报使用事件（开屏、对话时长等），供家属了解老人日常 |
+| **离线数字人渲染** | 数字人形象的 2D 渲染完全在本地完成（NCNN 推理 + OpenGL），无需 GPU 服务器 |
+| **低门槛部署** | 老人无需任何操作，设备预装即可使用；家属扫码绑定即可监护 |
+
+### 用户角色
+
+| 角色 | 使用场景 |
+|------|---------|
+| **老人**（设备端） | 与数字人语音聊天、听取家人留言、口述回复 |
+| **家属**（Web 端） | 发送留言、查看老人使用动态、接收推送通知 |
+
+---
+
+## 设计思路
+
+### 1. 端云分离、各司其职
+
+- **端侧（本项目）**：承担数字人渲染（完全离线）+ 实时 AI 对话（通过 Qwen Omni-Realtime 云端模型）+ 家庭消息同步
+- **云侧（Monitor Web）**：承担家属身份管理、设备绑定、消息存储与分发、使用数据可视化
+
+### 2. AI 对话与数字人解耦
+
+DUIX SDK 仅负责"给定 PCM 音频 → 渲染对应口型/表情动画"，不关心音频来源。音频可来自：
+- Qwen Omni-Realtime 模型的实时语音回复
+- 本地 TTS 合成
+- 任意 WAV 文件
+
+这种解耦使得 AI 模型可独立升级，不影响渲染管线。
+
+### 3. 工具调用实现双向留言
+
+通过 Qwen 的 Function Calling 能力，在 AI 对话中注册两个工具函数：
+- `leave_message_for_family`：老人口述 → AI 提取内容 → 调用工具 → 消息存入云端
+- `fetch_family_messages`：老人想听留言 → AI 调用工具 → 拉取消息 → 口播转达
+
+AI 自主判断何时调用工具，无需老人学习任何操作指令。
+
+### 4. 面向老人的交互设计
+
+- **语义 VAD**：采用 `semantic_vad` 而非传统声学 VAD，能过滤电视背景音、咳嗽等噪音，仅在老人真正说话时响应
+- **主动播报**：对话开始时若有未读家人留言，AI 会主动口播转达，无需老人主动查询
+- **温和人设**：系统 Prompt 设定为"面向中国老年群体的心理陪伴 AI"，使用温和、简短、生活化的中文回复
+
+---
+
+## 技术架构
+
+### 系统全景
+
+```
+┌──────────────────────────────────────────────────────────────────┐
+│                        Android 设备端                             │
+│                                                                  │
+│  ┌─────────────┐    PCM音频    ┌──────────────────────────────┐  │
+│  │ Qwen Omni   │ ────────────→ │        DUIX SDK              │  │
+│  │ Realtime    │               │ ┌────────────────────────┐   │  │
+│  │ (WebSocket) │    AI回复语音  │ │ NCNN 推理引擎          │   │  │
+│  │             │               │ │ (口型/表情/动作 推理)   │   │  │
+│  └──────┬──────┘               │ └───────────┬────────────┘   │  │
+│         │                      │             │ ImageFrame     │  │
+│  语音   │ 文本               │ ┌───────────▼────────────┐   │  │
+│  输入   │ +语音               │ │ OpenGL ES 2.0 渲染     │   │  │
+│         │                      │ │ (DUIXRenderer)         │   │  │
+│  ┌──────▼──────┐               │ └───────────┬────────────┘   │  │
+│  │  麦克风录音  │               │             │               │  │
+│  │  (16kHz PCM) │               │ ┌───────────▼────────────┐   │  │
+│  └─────────────┘               │ │ GLTextureView (UI)     │   │  │
+│                                └──────────────────────────────┘  │
+│                                                                  │
+│  ┌──────────────────────────────────────────────────────────┐    │
+│  │              FamilyBridgeService (HTTP)                   │    │
+│  │  • 设备注册 → POST /api/device/register                   │    │
+│  │  • 拉取留言 → GET  /api/device/messages                   │    │
+│  │  • 标记已读 → POST /api/device/messages/read              │    │
+│  │  • 上报事件 → POST /api/device/usage                      │    │
+│  │  • 对话存档 → POST /api/device/conversations              │    │
+│  │  • 工具记录 → POST /api/device/tool-call                  │    │
+│  └──────────────────────────────┬───────────────────────────┘    │
+└─────────────────────────────────┼────────────────────────────────┘
+                                  │ HTTPS
+                                  ▼
+┌──────────────────────────────────────────────────────────────────┐
+│           digital-human-monitor-v2 (Next.js 云端)                │
+│                                                                  │
+│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────────┐    │
+│  │ 设备管理  │  │ 消息存储  │  │ 行为分析  │  │ Web Push 推送 │    │
+│  └──────────┘  └──────────┘  └──────────┘  └──────────────┘    │
+│                       │                                          │
+│                       ▼                                          │
+│              ┌──────────────┐                                    │
+│              │  PostgreSQL   │                                    │
+│              └──────────────┘                                    │
+└──────────────────────────────────────────────────────────────────┘
 ```

-3. In the module's `build.gradle`, add the dependency:
+### 技术栈

-```gradle
-dependencies {
-    api project(":duix-sdk")
-}
+| 层级 | 技术 |
+|------|------|
+| **数字人渲染** | NCNN (端侧推理) + OpenCV 4.6 + OpenGL ES 2.0 + ONNX Runtime |
+| **AI 对话** | 千问 Qwen 3.5 Omni Plus Realtime (WebSocket) |
+| **音频处理** | AudioRecord (16kHz) → PCM → Base64 → WebSocket；AudioTrack 播放 |
+| **网络通信** | OkHttp 4 (HTTP API) + OkHttp WebSocket (Qwen Realtime) |
+| **本地存储** | SharedPreferences (设备 UUID、配置) |
+| **构建系统** | Gradle 8.13.2 + AGP 8.9.1 + JDK 17 (Temurin) |
+| **语言** | Kotlin (应用层) + Java (SDK 层) + C++ (Native 推理层) |
+
+---
+
+## 模块说明
+
+### 项目结构
+
+```
+dh_aigc_android/
+├── duix-sdk/           # 核心 2D 数字人渲染 SDK（库模块）
+│   ├── src/main/java/  # Java SDK 接口层
+│   │   └── ai/guiji/duix/
+│   │       ├── sdk/client/         # DUIX 主控、渲染、回调、工具类
+│   │       └── duix/DuixNcnn.java  # JNI 接口
+│   └── src/main/cpp/   # C++ Native 推理引擎
+│       ├── dhcore/     # 核心数字人推理
+│       ├── dhmfcc/     # 音频 MFCC 特征提取
+│       ├── dhunet/     # 口型网络模型
+│       └── third/      # 预编译依赖 (NCNN, OpenCV, ONNX, libjpeg)
+│
+├── test/               # 演示应用（应用模块）
+│   └── src/main/java/ai/guiji/duix/test/
+│       ├── ui/activity/
+│       │   ├── MainActivity.kt        # 主页：模型选择与下载管理
+│       │   ├── CallActivity.kt        # 核心：数字人渲染 + AI 对话 + 留言桥梁
+│       │   ├── SettingsActivity.kt     # 设备设置
+│       │   └── VoiceCloneSetupActivity.kt  # 声音克隆配置
+│       ├── service/
+│       │   ├── FamilyBridgeService.kt  # 家庭监护 API 客户端
+│       │   ├── OmniRealtimeClient.kt   # Qwen Omni WebSocket 客户端
+│       │   ├── QwenTtsRealtimeClient.kt # Qwen TTS 客户端
+│       │   ├── VoiceCloneService.kt    # 声音克隆服务
+│       │   └── DashScopeConfig.java    # API 配置常量
+│       ├── audio/
+│       │   ├── AudioResampler.java     # PCM 重采样
+│       │   └── StreamingAudioRecorder.java  # 实时麦克风录音
+│       └── camera/
+│           └── CameraFrameCapture.java # 摄像头帧采集
+│
+└── qwen-docs/          # 千问 Omni-Realtime API 参考文档
+    ├── main.md         # 协议总览
+    ├── client_events.md    # 客户端事件定义
+    ├── server_events.md    # 服务端事件定义
+    └── function_calling.md # 工具调用机制
 ```

-### 3.2 AAR Reference (Optional)
+### duix-sdk — 数字人渲染引擎

-1. Place the compiled `duix-sdk-release.aar` module into the `libs/` directory.
-2. Add the dependency:
+| 类 | 职责 |
+|----|------|
+| `DUIX` | 主控对象：管理初始化、渲染循环、音频驱动、生命周期 |
+| `RenderThread` | 独立渲染线程：逐帧推理 + 音频同步 |
+| `DUIXRenderer` | OpenGL ES 2.0 渲染器：将 ImageFrame 渲染到 TextureView |
+| `DUIXTextureView` | 自定义 TextureView：提供 GL 渲染表面 |
+| `VirtualModelUtil` | 模型资源管理：下载、校验、解压 |
+| `DuixNcnn` | JNI 桥接：调用 C++ 层 NCNN 推理 |

-```gradle
-dependencies {
-    api fileTree(include: ['*.jar', '*.aar'], dir: 'libs')
-}
+**渲染流水线**：
+```
+pushPcm(音频) → MFCC特征提取 → NCNN口型网络推理
+→ 获取ImageFrame(图像+掩码) → OpenGL混合渲染 → TextureView显示
+```
+
+### test — 应用集成层
+
+| 类 | 职责 |
+|----|------|
+| `CallActivity` | 整合 DUIX 渲染 + Qwen 对话 + 留言桥梁的核心页面 |
+| `OmniRealtimeClient` | Qwen Omni-Realtime WebSocket 客户端，处理实时语音对话 |
+| `FamilyBridgeService` | 家庭监护云 API 客户端，负责设备注册、消息同步、事件上报 |
+| `MainActivity` | 模型选择与下载管理入口 |
+
+---
+
+## 全链路数据流
+
+### 1. AI 实时对话
+
+```
+老人说话 → 麦克风录音 (16kHz PCM)
+→ Base64 编码 → input_audio_buffer.append (WebSocket)
+→ Qwen Omni 模型推理
+→ response.audio.delta (Base64 PCM 回传)
+→ 解码为 PCM → DUIX.pushPcm()
+→ NCNN 推理口型动画 → OpenGL 渲染
+→ 同时 AudioTrack 播放语音
+```
+
+### 2. 留言传递（老人 → 家属）
+
+```
+老人对 AI 说 "帮我给儿子带个话，说我挺好的"
+→ Qwen 识别意图 → 触发 leave_message_for_family 工具
+→ 提取 recipientRelation="儿子", message="我挺好的"
+→ FamilyBridgeService → POST /api/device/conversations
+→ Monitor Web 存入数据库
+→ Web Push 通知家属
+→ AI 回复老人 "好的，已经帮您告诉儿子了"
+```
+
+### 3. 留言传递（家属 → 老人）
+
+```
+家属在 Web 端发送留言 → 存入 FamilyMessage 表
+→ 老人设备启动 AI 对话
+→ FamilyBridgeService 拉取未读消息
+→ 注入 AI 系统 Prompt: "家里人刚留了新留言，请主动转告..."
+→ AI 在第一轮回复中口播: "您儿子给您捎了个话——记得吃药"
+```
+
+### 4. 使用行为上报
+
+```
+设备事件 (APP_OPEN / AI_SESSION_STARTED / TOOL_CALLED / ...)
+→ FamilyBridgeService → POST /api/device/usage
+→ Monitor Web 存入 UsageEvent 表
+→ 家属在"陪伴动态"页面查看趋势图
 ```

 ---

-## 4. Integration Requirements
+## 通信协议

-| Item           | Description                                                     |
-|----------------|-----------------------------------------------------------------|
-| System         | Supports Android 10+ systems.                                    |
-| CPU Architecture | armeabi-v7a, arm64-v8a                                           |
-| Hardware Requirements | Device CPU with 8 or more cores (Snapdragon 8 Gen 2), 8GB or more memory, available storage space of 1GB or more |
-| Network        | None (Fully local operation)                                    |
-| Development IDE | Android Studio Giraffe 2022.3.1 Patch 2                         |
-| Memory Requirements | Minimum 800MB memory available for the digital human          |
+### Qwen Omni-Realtime WebSocket
+
+| 方向 | 事件 | 用途 |
+|------|------|------|
+| C→S | `session.update` | 设置模态(text+audio)、语音、VAD、系统 Prompt、工具列表 |
+| C→S | `input_audio_buffer.append` | 推送用户麦克风 PCM (Base64) |
+| S→C | `response.audio.delta` | 模型语音回复增量块 (Base64 PCM) |
+| S→C | `response.text.done` | 模型文本回复完成 |
+| S→C | `response.function_call_arguments.done` | 工具调用请求 |
+| C→S | `conversation.item.create` | 回传工具执行结果 |
+| C→S | `response.create` | 触发模型基于工具结果生成最终回复 |
+
+### 注册的 AI 工具
+
+| 工具名 | 参数 | 用途 |
+|--------|------|------|
+| `leave_message_for_family` | `recipientRelation` (称谓), `message` (内容), `importance` (优先级) | 老人留言给家人 |
+| `fetch_family_messages` | `count` (数量 1-5) | 拉取家人留言 |
+
+### FamilyBridge HTTP API
+
+| 端点 | 方法 | 用途 |
+|------|------|------|
+| `/api/device/register` | POST | 设备注册，返回 `deviceUuid` + `bindUrl` |
+| `/api/device/messages` | GET | 拉取待送达消息 |
+| `/api/device/messages/read` | POST | 标记消息已读 |
+| `/api/device/usage` | POST | 上报使用事件 |
+| `/api/device/conversations` | POST | 上传对话轮次 |
+| `/api/device/tool-call` | POST | 记录工具调用日志 |

 ---

-## 5. Usage Flow Overview
+## 集成要求

-```mermaid
-graph TD
-A[Check Configuration and Models] --> B[Build DUIX Instance]
-B --> C[Call init to Initialize]
-C --> D[Display Avatar / Render]
-D --> E[PCM or WAV Audio Driving]
-E --> F[Playback Control & Motion Triggering]
-F --> G[Resource Release]
-```
+| 项目 | 要求 |
+|------|------|
+| 系统版本 | Android 10+ (API 24+) |
+| CPU 架构 | arm64-v8a, armeabi-v7a |
+| 硬件 | 8 核 CPU (骁龙 8 Gen2), 8GB+ 内存, 1GB+ 存储 |
+| 内存 | 数字人运行需 ≥ 800MB 可用内存 |
+| 网络 | 数字人渲染：无需网络；AI 对话 + 监护同步：需联网 |
+| 开发 IDE | Android Studio Giraffe 2022.3.1+ |
+| JDK | OpenJDK 17 (Temurin) |
+| Gradle | 8.13.2 (使用项目自带 wrapper) |

 ---

-## 6. Key Interfaces and Example Calls
+## SDK 接口文档

-### 6.1 Model Check and Download
+### 术语

-Before using the rendering service, ensure that the basic configuration and model files are synchronized to local storage. The SDK provides a simple demonstration of the model download and decompression process using `VirtualModelUtil`. If model download is slow or fails, developers can choose to cache the model package to their own storage service.
+| 术语 | 含义 |
+|------|------|
+| PCM | 16kHz 采样率、16bit 位深、Mono 单通道的原始音频流 |
+| WAV | 支持 PCM 编码的音频文件格式 |
+| RenderSink | 渲染数据接收接口，SDK 提供默认实现 |
+| DUIX | 数字人主控对象 |
+| ImageFrame | 包含 rawBuffer (图像) + maskBuffer (掩码) + 宽高的渲染帧 |

-> Function Definition: `ai.guiji.duix.sdk.client.VirtualModelUtil`
-
-```
-// Check if base configuration is downloaded
-boolean checkBaseConfig(Context context)
-
-// Check if the model is downloaded
-boolean checkModel(Context context, String name)
-
-// Base configuration download
-void baseConfigDownload(Context context, String url, ModelDownloadCallback callback)
-
-// Model download
-void modelDownload(Context context, String modelUrl, ModelDownloadCallback callback)
-```
-
-`ModelDownloadCallback` includes progress, completion, failure callbacks, etc., as defined in the SDK.
-
-```
-interface ModelDownloadCallback {
-    // Download progress
-    void onDownloadProgress(String url, long current, long total);
-    // Unzip progress
-    void onUnzipProgress(String url, long current, long total);
-    // Download and unzip complete
-    void onDownloadComplete(String url, File dir);
-    // Download and unzip failed
-    void onDownloadFail(String url, int code, String msg);
-}
-```
-
-**Call Example**:
+### 模型检查与下载

 ```kotlin
-if (!VirtualModelUtil.checkBaseConfig(mContext)){
-    VirtualModelUtil.baseConfigDownload(mContext, baseConfigUrl, callback)
-}
+// 检查基础配置
+VirtualModelUtil.checkBaseConfig(context): Boolean
+// 下载基础配置
+VirtualModelUtil.baseConfigDownload(context, url, callback)
+
+// 检查模型
+VirtualModelUtil.checkModel(context, modelUrl): Boolean
+// 下载模型
+VirtualModelUtil.modelDownload(context, modelUrl, callback)
 ```

+### 初始化与渲染
+
 ```kotlin
-if (!VirtualModelUtil.checkModel(mContext, modelUrl)){
-    VirtualModelUtil.modelDownload(mContext, modelUrl, callback)
-}
-```
-
---
-
-### 6.2 Initialization and Rendering Start
-
-In the `onCreate()` stage of the rendering page, build the DUIX object and call the init interface.
-
-> Function Definition: `ai.guiji.duix.sdk.client.DUIX`
-
-```
-// Build DUIX object
-public DUIX(Context context, String modelName, RenderSink sink, Callback callback)
-
-// Initialize DUIX service
-void init()
-```
-
-**DUIX Object Construction Explanation**:
-
-| Parameter     | Type      | Description                                                    |
-|---------------|-----------|----------------------------------------------------------------|
-| context       | Context   | System context                                                  |
-| modelName     | String    | Can pass the model download URL (if downloaded) or cached filename |
-| render        | RenderSink| Rendering data interface, SDK provides a default rendering component inheriting from this interface, or you can implement it yourself |
-| callback      | Callback  | Various callback events handled by the SDK                      |
-
-Where **Callback** is defined as: `ai.guiji.duix.sdk.client.Callback`
-
-```
-interface Callback {
-    void onEvent(String event, String msg, Object info);
-}
-```
-
-**Call Example**:
-
-```kotlin
-duix = DUIX(mContext, modelUrl, mDUIXRender) { event, msg, info ->
+// 构建 DUIX 实例
+val duix = DUIX(context, modelUrl, renderSink) { event, msg, info ->
    when (event) {
-        ai.guiji.duix.sdk.client.Constant.CALLBACK_EVENT_INIT_READY -> {
-            initOK()
-        }
-
-        ai.guiji.duix.sdk.client.Constant.CALLBACK_EVENT_INIT_ERROR -> {
-            initError()
-        }
-        // ...
+        Constant.CALLBACK_EVENT_INIT_READY -> { /* 初始化完成 */ }
+        Constant.CALLBACK_EVENT_INIT_ERROR -> { /* 初始化失败 */ }
+        Constant.CALLBACK_EVENT_PLAY_START -> { /* 播放开始 */ }
+        Constant.CALLBACK_EVENT_PLAY_END   -> { /* 播放结束 */ }
    }
 }
-// Asynchronous callback result
-duix?.init()
+duix.init()
 ```

-In the `init` callback, confirm the initialization result.
+### 音频驱动
+
+```kotlin
+// PCM 流式推送（用于实时 AI 语音）
+duix.pushPcm(pcmBytes)
+
+// WAV 文件播放
+duix.playWav(wavFilePath)
+```
+
+### 资源释放
+
+```kotlin
+// 在 Activity onDestroy 中调用
+duix.release()
+```

 ---

-### 6.3 Digital Human Avatar Display
+## 构建与运行

-Use the SDK-provided `DUIXRenderer` and `DUIXTextureView` to quickly implement rendering with transparency support. Alternatively, you can implement the `RenderSink` interface to customize the rendering logic.
+### 环境准备

-The **RenderSink** definition is as follows: `ai.guiji.duix.sdk.client.render.RenderSink`
+1. 安装 Android Studio Giraffe 2022.3.1+
+2. 配置 JDK 17：`File → Settings → Build → Gradle → Gradle JDK → 选择 JDK 17`
+3. 连接 Android 设备（需满足硬件要求）
+
+### 编译安装
+
+```bash
+# 使用 Gradle Wrapper 编译安装 test 应用
+./gradlew :test:installDebug
+```
+
+### 配置说明
+
+API 密钥和服务地址在 `DashScopeConfig.java` 中配置：

 ```java
-/**
- * Rendering pipeline, returns rendering data through this interface
- */
-public interface RenderSink {
-
-    // The frame's buffer data is arranged in BGR order
-    void onVideoFrame(ImageFrame imageFrame);
-
-}
-```
-
-**Call Example**:
-
-Use `DUIXRenderer` and `DUIXTextureView` to quickly implement rendering. These controls support transparency and can freely set the background and foreground.
-
-```kotlin
-override fun onCreate(savedInstanceState: Bundle?) {
-    super.onCreate(savedInstanceState)
-    // ...
-    mDUIXRender =
-        DUIXRenderer(
-            mContext,
-            binding.glTextureView
-        )
-
-    binding.glTextureView.setEGLContextClientVersion(GL_CONTEXT_VERSION)
-    binding.glTextureView.setEGLConfigChooser(8, 8, 8, 8, 16, 0) // Transparency
-    binding.glTextureView.isOpaque = false           // Transparency
-    binding.glTextureView.setRenderer(mDUIXRender)
-    binding.glTextureView.renderMode =
-        GLSurfaceView.RENDERMODE_WHEN_DIRTY      // Must be called after setting the renderer
-
-    duix = DUIX(mContext, modelUrl, mDUIXRender) { event, msg, _ ->
-    }
-    // ...
-}
+MONITOR_BASE_URL   // 监护 Web 服务地址
+REALTIME_BASE_URL  // Qwen Omni WebSocket 地址
+DEFAULT_API_KEY    // DashScope API Key
+AUDIO_SAMPLE_RATE  // 音频采样率 (24000)
 ```

 ---

-### 6.4 Broadcasting Control
+## 关联项目

-#### Use Streaming PCM to Drive Digital Human Broadcasting
-
-**PCM Format: 16kHz sample rate, single channel, 16-bit depth**
-
-> Function Definition: `ai.guiji.duix.sdk.client.DUIX`
-
-```
-// Notify service to start pushing audio
-void startPush()
-
-// Push PCM data
-void pushPcm(byte[] buffer)
-
-// Finish a segment of audio push (Call this after the audio push is complete, not after playback finishes)
-void stopPush()
-```
-
-`startPush`, `pushPcm`, and `stopPush` need to be called in pairs. `pushPcm` should not be too long. After pushing the entire audio, call `stopPush` to end the session. Use `startPush` again for the next audio.
-
-**The audio data between each startPush and stopPush segment should be at least 1 second (32000 bytes), otherwise the mouth shape driver cannot be triggered, and blank frames can be used to fill in.**
-
-**Call Example**:
-
-```kotlin
-val thread = Thread {
-            duix?.startPush()
-            val inputStream = assets.open("pcm/2.pcm")
-            val buffer = ByteArray(320)
-            var length = 0
-            while (inputStream.read(buffer).also { length = it } > 0){
-                val data = buffer.copyOfRange(0, length)
-                duix?.pushPcm(data)
-            }
-            duix?.stopPush()
-            inputStream.close()
-}
-thread.start()
-```
-
---
-
-### 6.5 Motion Control
-
-#### Play Specific Motion Interval
-
-The model supports new motion intervals marked in `SpecialAction.json`
-
-> Function Definition: `ai.guiji.duix.sdk.client.DUIX`
-
-```
-/**
- * Play specific motion interval
- * @param name The motion interval name, which can be obtained from @{ModelInfo.getSilenceRegion()} after init callback
- * @param now Whether to play immediately: true: play now; false: wait for current silent or motion interval to finish
- */
-void startMotion(String name, boolean now)
-```
-
-**Call Example**:
-
-```kotlin
-duix?.startMotion("Greeting", true)
-```
-
-#### Randomly Play Motion Interval
-
-> Function Definition: `ai.guiji.duix.sdk.client.DUIX`
-
-```
-/**
- * Randomly play a motion interval
- * @param now Whether to play immediately: true: play now; false: wait for current silent or motion interval to finish
- */
-void startRandomMotion(boolean now);
-```
-
-**Call Example**:
-
-```kotlin
-duix?.startRandomMotion(true)
-```
-
---
-
-## 7. Proguard Configuration
-
-If using obfuscation, add the following in `proguard-rules.pro`:
-
-```proguard
-keep class ai.guiji.duix.DuixNcnn{*; }
-```
-
---
-
-## 8. Precautions
-
-1. Ensure that the base configuration file and model are downloaded to the specified location before driving rendering initialization.
-2. PCM audio should not be too long, as PCM buffers are cached in memory; long audio streams may cause memory overflow.
-3. To replace the preview model, modify the `modelUrl` value in `MainActivity.kt` and use the SDK's built-in file download and decompression management to obtain the complete model files.
-4. Audio driving format: 16kHz sample rate, single channel, 16-bit depth.
-5. Insufficient device performance may result in the audio feature extraction speed not matching the playback speed. You can use `duix?.setReporter()` to monitor frame rendering information.
-
---
-
-## 9. FAQ and Troubleshooting Guide
-
-| Issue                          | Possible Cause               | Solution                     |
-|---------------------------------|------------------------------|------------------------------|
-| init callback failed            | Model path error or model not downloaded | Use `checkModel` to check model status |
-| Rendering black screen          | EGL configuration or texture view error | Use SDK-provided example settings |
-| No PCM playback effect          | Incorrect format or `startPush` not called | Ensure audio format is correct and call push method |
-| Model download slow             | Unstable network or restricted CDN | Support self-hosted model file storage service |
-
---
-
-## 10. Version History
-
-**<a>4.0.1</a>**
-
-1. Supports PCM audio stream driving the digital human, improving audio playback response speed.
-2. Optimized motion interval playback, allowing specific motion intervals based on model configuration.
-3. Custom audio player, removed Exoplayer playback dependency.
-4. Provided simplified model download synchronization management tools.
-5. The audio data between each startPush and stopPush segment should be at least 1 second (32000 bytes), otherwise the mouth shape driver cannot be triggered, and blank frames can be used to fill in.
-
-**<a>3.0.5</a>**
-
-```text
-1. Updated arm32 CPU libonnxruntime.so version to fix compatibility issues.
-2. Modified motion interval playback function, supports random and sequential playback, requires manual call to stop playback to return to silent interval.
-```
-
-**<a>3.0.4</a>**
-
-```text
-1. Fixed model display issue due to low float precision on some devices.
-```
-
-**<a>3.0.3</a>**
-
-```text
-1. Optimized local rendering.
-```
-
-## 11. 🔗 Open-source Dependencies
-
-| Module                                   | Description                    |
-|------------------------------------------|--------------------------------|
-| [onnx](https://github.com/onnx/onnx)     | General AI model standard format |
-| [ncnn](https://github.com/Tencent/ncnn)  | High-performance neural network computing framework (Tencent) |
-
---
-
-For more help, please contact the technical support team.
+| 项目 | 角色 | 仓库 |
+|------|------|------|
+| **dh_aigc_android**（本项目） | 老人端：AI 数字人陪护 + 设备行为上报 | 当前仓库 |
+| **digital-human-monitor-v2** | 家属端：远程监护 Web 面板 + 消息管理 + 推送通知 | [digital-human-monitor-v2](../digital-human-monitor-v2/) |