sandrohanea/whisper.net

Whisper.net. Speech to text made simple using Whisper Models

What it solves

Whisper.net 为 OpenAI 的 Whisper 提供 .NET 绑定，允许开发者直接在 C# 和 .NET 应用程序中集成高质量的语音转文本和翻译功能，而无需编写原生 C++ 代码。

How it works

它作为 whisper.cpp 的封装层，利用 Ggml 模型进行推理。该项目提供了一个灵活的运行时系统，可以根据平台和硬件自动选择最佳的可用硬件加速，支持 CPU、NVIDIA CUDA (12 和 13)、Intel OpenVino、Apple CoreML、Vulkan 和 Vulkan。

Who it’s for

希望在 Windows、Linu x, macOS, iOS, Android, 和 WebAssembly 等各种平台上实现本地、离线语音识别和翻译的 .NET 开发者。

Highlights

  • Multi-platform support: Works across desktop, mobile, and web (WASM).
  • Hardware acceleration: Built-in support for CUDA, CoreML, OpenVino, and Vulkan.
  • Automatic runtime selection: Transparently falls back from GPU to CPU based on available drivers and hardware.
  • Easy model management: Includes a Hugging Face downloader for Ggml models and Silero VAD (Voice Activity Detection) models.
  • Flexible loading: Models can be loaded from files, streams, or buffers.