whisper.cpp
FreeWhisper reimplemented in C++ — runs on CPU, no Python required
A C++ inference implementation of OpenAI Whisper with no Python dependency, capable of running full models on CPU. No GUI; it ships as a command line tool and a library, suited to scripts and your own programs. The most actively developed of the common options.
Why we picked it
whisper.cpp is the least painful way to wire local transcription into your own workflow. 52,897 stars, MIT, and the latest v1.9.2 shipped on 4 August 2026 — the most frequently updated of this group.
Its practical advantage is needing no Python environment. Dependency setup is where most people get stuck; here you compile it or install via a package manager and it runs. CPU-only works, just slower.
There is no GUI. For point-and-click use see Vibe or Buzz; for subtitles with speaker labels see whisperX.
One thing that matters more than the tool choice: accuracy depends on the model size you can run. tiny is 75 MiB but error-prone; large-v3 needs 2.9 GiB, with the quantized build at 1.1 GiB as the sensible middle ground.