🌍 A Cross-Platform On-Device Offline AI Voice & Translation App / 一款跨平台的端侧离线 AI 语音与翻译应用
⚠️ Important Notice: VoiceHub relies on local open-source AI models to function. Before running the app, you must download and configure the required models. Please refer to the Models Setup Guide for detailed instructions.
VoiceHub is a powerful cross-platform Flutter application providing dedicated iOS and macOS apps. It integrates advanced on-device AI models for Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and Neural Machine Translation (NMT), delivering a complete voice processing pipeline right on your device.
llama.cpp inference framework, allowing Large Language Models (LLMs) to run directly on the device's GPU for blazing fast translations.Note: Replace the placeholders below with actual application screenshots.
| Main Screen | Cascade Translation | Dual Dialogue |
|---|---|---|
![]() | ![]() | ![]() |
| Simultaneous Translation | File Translation | History Records |
|---|---|---|
![]() | ![]() | ![]() |
VoiceHub provides extensive support for various open-source models:
transducer architecture models released by sherpa-onnx.vits architecture models released by sherpa-onnx.opus_mt models to ONNX format.gguf format. We highly recommend using the hy-mt2-1.8B model for translation tasks.During continuous audio recording, VoiceHub primarily relies on the silero-vad model for Voice Activity Detection (VAD) and automatic sentence segmentation.
0.5, with minimum speech and silence durations both set to 0.3 seconds (300ms) for snappy, responsive turns.0.5 seconds (500ms) to allow longer continuous speech blocks before translating.VoiceHub's robust architecture is built upon several excellent open-source projects:
n_ctx tuning), Q4_K_M/Q8 quantized SpeechLLM support, and llama translation KV cache cleanup.Please refer to the following documentation in the @docs directory for build instructions, model setup, and platform-specific guides:
⚠️ 重要提示:VoiceHub 的所有核心功能均依赖于本地开源 AI 模型。在运行应用之前,您必须先下载并配置相关模型。请务必仔细参考 模型配置指南 进行操作。
VoiceHub 是一款强大的跨平台 Flutter 应用,目前已提供原生的 iOS 和 macOS 应用版本。它集成了先进的端侧 AI 模型,提供自动语音识别 (ASR)、文字转语音 (TTS) 以及神经机器翻译 (NMT) 等完整的语音处理能力。
llama.cpp 推理框架,VoiceHub 能够将大语言模型 (LLM) 直接运行在设备的 GPU 上,从而大幅加速翻译推理速度。注:请将下方的占位图替换为实际的应用截图。
| 主界面 | 级联翻译 | 双人对话 |
|---|---|---|
![]() | ![]() | ![]() |
| 同声传译 | 文件翻译 | 历史记录 |
|---|---|---|
![]() | ![]() | ![]() |
VoiceHub 为多种开源模型提供深度的本地推理支持:
sherpa-onnx 发布的 transducer 架构模型。sherpa-onnx 发布的 vits 架构模型。opus_mt 模型导出为 ONNX 格式后使用。gguf 格式的开源大模型。针对翻译场景,强烈推荐使用腾讯的 hy-mt2-1.8B 翻译大模型。在持续收音过程中,应用主要依赖于自带的 silero-vad 模型来进行实时的语音端点检测与自动断句。
0.5,最小静音 (minSilence) 和最小语音 (minSpeech) 时长均为 0.3 秒 (300ms),以实现极速响应。0.5 秒 (500ms)。VoiceHub 的强大功能离不开以下优秀的开源项目:
n_ctx 下调防 OOM)、支持 Q4_K_M/Q8 量化 SpeechLLM,以及翻译后立刻释放 llama KV cache。关于环境搭建、编译打包和模型配置,请参考项目 @docs 目录下的详细教程:
52 commits
C++
50.6%
C
12.7%
Python
10.3%
Cuda
3.5%
Dart
3.0%
TypeScript
2.5%
Shell
1.9%
HTML
1.8%
Svelte
1.4%
JavaScript
1.4%
CMake
1.4%
Kotlin
1.1%
Rust
1.1%
🌍 A Cross-Platform On-Device Offline AI Voice & Translation App / 一款跨平台的端侧离线 AI 语音与翻译应用
⚠️ Important Notice: VoiceHub relies on local open-source AI models to function. Before running the app, you must download and configure the required models. Please refer to the Models Setup Guide for detailed instructions.
VoiceHub is a powerful cross-platform Flutter application providing dedicated iOS and macOS apps. It integrates advanced on-device AI models for Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and Neural Machine Translation (NMT), delivering a complete voice processing pipeline right on your device.
llama.cpp inference framework, allowing Large Language Models (LLMs) to run directly on the device's GPU for blazing fast translations.Note: Replace the placeholders below with actual application screenshots.
| Main Screen | Cascade Translation | Dual Dialogue |
|---|---|---|
![]() | ![]() | ![]() |
| Simultaneous Translation | File Translation | History Records |
|---|---|---|
![]() | ![]() | ![]() |
VoiceHub provides extensive support for various open-source models:
transducer architecture models released by sherpa-onnx.vits architecture models released by sherpa-onnx.opus_mt models to ONNX format.gguf format. We highly recommend using the hy-mt2-1.8B model for translation tasks.During continuous audio recording, VoiceHub primarily relies on the silero-vad model for Voice Activity Detection (VAD) and automatic sentence segmentation.
0.5, with minimum speech and silence durations both set to 0.3 seconds (300ms) for snappy, responsive turns.0.5 seconds (500ms) to allow longer continuous speech blocks before translating.VoiceHub's robust architecture is built upon several excellent open-source projects:
n_ctx tuning), Q4_K_M/Q8 quantized SpeechLLM support, and llama translation KV cache cleanup.Please refer to the following documentation in the @docs directory for build instructions, model setup, and platform-specific guides:
⚠️ 重要提示:VoiceHub 的所有核心功能均依赖于本地开源 AI 模型。在运行应用之前,您必须先下载并配置相关模型。请务必仔细参考 模型配置指南 进行操作。
VoiceHub 是一款强大的跨平台 Flutter 应用,目前已提供原生的 iOS 和 macOS 应用版本。它集成了先进的端侧 AI 模型,提供自动语音识别 (ASR)、文字转语音 (TTS) 以及神经机器翻译 (NMT) 等完整的语音处理能力。
llama.cpp 推理框架,VoiceHub 能够将大语言模型 (LLM) 直接运行在设备的 GPU 上,从而大幅加速翻译推理速度。注:请将下方的占位图替换为实际的应用截图。
| 主界面 | 级联翻译 | 双人对话 |
|---|---|---|
![]() | ![]() | ![]() |
| 同声传译 | 文件翻译 | 历史记录 |
|---|---|---|
![]() | ![]() | ![]() |
VoiceHub 为多种开源模型提供深度的本地推理支持:
sherpa-onnx 发布的 transducer 架构模型。sherpa-onnx 发布的 vits 架构模型。opus_mt 模型导出为 ONNX 格式后使用。gguf 格式的开源大模型。针对翻译场景,强烈推荐使用腾讯的 hy-mt2-1.8B 翻译大模型。在持续收音过程中,应用主要依赖于自带的 silero-vad 模型来进行实时的语音端点检测与自动断句。
0.5,最小静音 (minSilence) 和最小语音 (minSpeech) 时长均为 0.3 秒 (300ms),以实现极速响应。0.5 秒 (500ms)。VoiceHub 的强大功能离不开以下优秀的开源项目:
n_ctx 下调防 OOM)、支持 Q4_K_M/Q8 量化 SpeechLLM,以及翻译后立刻释放 llama KV cache。关于环境搭建、编译打包和模型配置,请参考项目 @docs 目录下的详细教程:
52 commits
C++
50.6%
C
12.7%
Python
10.3%
Cuda
3.5%
Dart
3.0%
TypeScript
2.5%
Shell
1.9%
HTML
1.8%
Svelte
1.4%
JavaScript
1.4%
CMake
1.4%
Kotlin
1.1%
Rust
1.1%