sherpa_asr_sdk 1.0.0

Offline speech recognition SDK for Flutter using Sherpa-onnx. Supports real-time streaming ASR with automatic model download.

Sherpa ASR SDK

English | 中文

---

English

Offline speech recognition SDK for Flutter using Sherpa-onnx. Supports real-time streaming ASR with automatic model download.

Features

• 🎤 Offline Recognition - No internet required, powered by Sherpa-onnx
• 🔄 Real-time Streaming - Get results as you speak
• 📦 Auto Download - Models downloaded automatically on first use
• 🔧 Model Management - Download and switch between models
• 📱 Cross Platform - iOS and Android support

Installation

Add to your pubspec.yaml:

dependencies:
  sherpa_asr_sdk: ^1.0.0

Quick Start

import 'package:sherpa_asr_sdk/sherpa_asr_sdk.dart';

// 1. Initialize (call once at app startup)
AsrSdk.setLogger(DefaultAsrLogger());
final success = await AsrSdk.initialize(
  onProgress: (p) => print('Loading: ${(p * 100).toInt()}%'),
);

// 2. Start service (when entering a page)
await AsrSdk.start();

// 3. Recognize speech
AsrSdk.recognize().listen((text) {
  print('Recognized: $text');
});

// 4. Stop recognition
await AsrSdk.stopRecognition();

// 5. Stop service (when leaving page)
await AsrSdk.stop();

// 6. Dispose (when app exits)
await AsrSdk.dispose();

API Reference

Lifecycle Methods

Method	Description
initialize()	Initialize SDK (call once at startup)
start()	Start service (create recorder)
stop()	Stop service (destroy recorder)
dispose()	Release all resources

Recognition Methods

Method	Description
recognize()	Start recognition, returns Stream<String>
stopRecognition()	Stop current recognition
cancelRecognition()	Cancel current recognition

State Properties

Property	Description
isInitialized	SDK initialized
isStarted	Service started
isListening	Recognition in progress
state	Current SDK state
stateStream	Stream of state changes

Model Management

final manager = SherpaModelsManager.instance;

// Check models
final hasModel = await manager.hasStreamingBilingualModel();

// Download model
await manager.downloadStreamingBilingualModels(
  onProgress: (p) => print('Download: ${(p * 100).toInt()}%'),
);

Model Types

Type	Description	Size
Streaming Bilingual	Chinese-English	~30MB
Base	Chinese only	~15MB
Advanced	Higher quality	~50MB

Platform Setup

iOS

Add to ios/Runner/Info.plist:

<key>NSMicrophoneUsageDescription</key>
<string>This app needs microphone access for speech recognition</string>

Android

Add to android/app/src/main/AndroidManifest.xml:

<uses-permission android:name="android.permission.RECORD_AUDIO" />
<uses-permission android:name="android.permission.INTERNET" />

---

中文

基于 Sherpa-onnx 的 Flutter 离线语音识别 SDK。支持实时流式识别，自动下载模型。

功能特性

• 🎤 离线识别 - 无需联网，基于 Sherpa-onnx
• 🔄 实时流式 - 边说边识别，即时返回结果
• 📦 自动下载 - 首次使用时自动下载模型
• 🔧 模型管理 - 支持下载和切换不同模型
• 📱 跨平台 - 支持 iOS 和 Android

安装

在 pubspec.yaml 中添加：

dependencies:
  sherpa_asr_sdk: ^1.0.0

快速开始

import 'package:sherpa_asr_sdk/sherpa_asr_sdk.dart';

// 1. 初始化（应用启动时调用一次）
AsrSdk.setLogger(DefaultAsrLogger());
final success = await AsrSdk.initialize(
  onProgress: (p) => print('加载中: ${(p * 100).toInt()}%'),
);

// 2. 启动服务（进入页面时调用）
await AsrSdk.start();

// 3. 开始识别
AsrSdk.recognize().listen((text) {
  print('识别结果: $text');
});

// 4. 停止识别
await AsrSdk.stopRecognition();

// 5. 停止服务（离开页面时调用）
await AsrSdk.stop();

// 6. 释放资源（应用退出时调用）
await AsrSdk.dispose();

API 参考

生命周期方法

方法	描述
initialize()	初始化 SDK（启动时调用一次）
start()	启动服务（创建录音器）
stop()	停止服务（销毁录音器）
dispose()	释放所有资源

识别方法

方法	描述
recognize()	开始识别，返回 Stream<String>
stopRecognition()	停止当前识别
cancelRecognition()	取消当前识别

状态属性

属性	描述
isInitialized	SDK 是否已初始化
isStarted	服务是否已启动
isListening	是否正在识别
state	当前 SDK 状态
stateStream	状态变化流

模型管理

final manager = SherpaModelsManager.instance;

// 检查模型
final hasModel = await manager.hasStreamingBilingualModel();

// 下载模型
await manager.downloadStreamingBilingualModels(
  onProgress: (p) => print('下载中: ${(p * 100).toInt()}%'),
);

模型类型

类型	描述	大小
流式中英双语	中英文混合识别	约 30MB
基础中文	纯中文识别	约 15MB
高级模型	更高质量	约 50MB

平台配置

iOS

在 ios/Runner/Info.plist 中添加：

<key>NSMicrophoneUsageDescription</key>
<string>此应用需要麦克风权限以进行语音识别</string>

Android

在 android/app/src/main/AndroidManifest.xml 中添加：

<uses-permission android:name="android.permission.RECORD_AUDIO" />
<uses-permission android:name="android.permission.INTERNET" />

---

Example / 示例

See the example/ directory for a complete sample app.

查看 example/ 目录获取完整示例应用。

License / 许可证

MIT License