Transcribe.cpp: open-source local offline AI speech recognition for 60+ model families

What happened

Transcribe.cpp, introduced on July 19, 2026 on Hacker News (566 points), is a new C/C++ library for offline AI speech recognition — a direct successor to whisper.cpp with expanded model support and platform acceleration.

Context and impact

The library responds to growing demand for local AI audio processing without cloud API dependencies (OpenAI Whisper API, AssemblyAI, Google Cloud Speech). The goal is straightforward integration in desktop and mobile apps while preserving full data privacy.

Details

• Supports 60+ transcription models from 16 model families
• Hardware acceleration: Vulkan, Metal, CUDA, TinyBLAS
• Language bindings: Python, JavaScript/TypeScript, Rust, Objective-C/Swift
• Numerical validation of outputs against reference implementations
• Supports both streaming and batch transcription
• Fully local operation: no data leaves the device