Transcribe.cpp

A ggml-based speech-to-text library covering 16 ASR families and 60+ models, largely compatible with whisper.cpp.

What it is

Transcribe.cpp is a ggml-based speech-to-text inference library for cross-platform local transcription, maintained by the creator of the Handy speech-to-text app. It builds on ggml for its community support and distribution story for local inference.

Capabilities

It supports 16 ASR families (60+ models, with more coming), hardware acceleration via Vulkan, Metal, CUDA and TinyBLAS, plus streaming and batch transcription. Models are numerically validated with Word Error Rate testing against reference implementations.

It is largely compatible with whisper.cpp, and ships official bindings for Python, JavaScript/TypeScript, Rust and Objective-C/Swift.

Performance

The developer notes an RK3566 can run models via transcribe.cpp faster than real time on its “anemic CPU” — a useful signal for edge and on-device transcription.