简介

本项目展示了如何在 Flutter 应用中使用 Sherpa-ONNX 进行实时语音识别。它基于 Sherpa-ONNX 项目的官方 streaming_asr 示例，但增加了额外的功能和优化。

主要特点

• 按需加载模型：语音识别模型可在需要时动态下载，显著减少应用初始大小
• 实时识别：以最小延迟处理语音
• 跨平台支持：同时支持 iOS 和 Android 设备

实现细节

本示例展示了如何：

• 在 Flutter 环境中初始化 Sherpa-ONNX 运行时
• 实时捕获和处理音频流
• 处理模型下载和缓存
• 通过用户友好的界面显示识别结果

应用截图 Screenshots

首次启动时的模型下载提示界面 Initial Model Download Prompt

模型下载进度展示 Model Download Progress

模型解压缩过程界面 Model Extraction Process

模型准备完成界面 Model Preparation Complete

实时语音识别运行界面 Real-time Speech Recognition

参考资料

• Sherpa-ONNX
• Streaming ASR Example

Introduction

This project demonstrates real-time speech recognition in Flutter applications using the Sherpa-ONNX runtime. It's based on the official streaming_asr example from the Sherpa-ONNX project but extends it with additional features and optimizations.

Key Features

• On-demand Model Loading: Speech recognition models are downloaded dynamically when needed, significantly reducing the app's initial size
• Real-time Recognition: Process speech as it's being spoken with minimal latency
• Cross-platform Support: Works on both iOS and Android devices

Implementation Details

The example showcases how to:

• Initialize the Sherpa-ONNX runtime in a Flutter environment
• Capture and process audio streams in real-time
• Handle model downloading and caching
• Display recognition results with a user-friendly interface

References

• Sherpa-ONNX GitHub Repository
• Original Streaming ASR Example