Sunday, July 19, 2026

transcribe.cpp

★1,674

One ggml STT engine for sixteen-plus model families, not just Whisper.

transcribe.cpp

Why it matters

Local speech stacks still fragment across Whisper forks and one-off exporters. transcribe.cpp is a C/C++ GGUF runtime on ggml with Metal, Vulkan, CUDA, and CPU paths, covering Parakeet, Canary, Moonshine, Qwen3-ASR, Voxtral, Nemotron streaming, and even MOSS Transcribe-Diarize. Models are numerically checked against reference implementations. The practical win is one local inference layer for modern STT instead of a pile of incompatible runners.

speechggmllocal-inference
Source ↗

Via github

X