Whisper-Ctranslate2

whisper-ctranslate2 is an open-source command-line speech-to-text tool that reimplements OpenAI's Whisper on top of CTranslate2 and Faster-Whisper, aiming for the same accuracy as the original at up to 4x the speed (up to 16x with batched inference) while using less memory. It's designed as a drop-in replacement for OpenAI's own Whisper CLI, supporting the same command-line options plus CTranslate2-specific extras like batched inference, quantization control (int8/float16/etc.), loading a custom or fine-tuned Whisper model from a local directory, speaker diarization, a voice activity detection (VAD) filter to improve quality on noisy audio, live transcription from a microphone, and color-coded confidence visualization in its output. It installs via pip, or as a prebuilt Docker image that ships with the small, medium, and large-v2 models already included, and runs on CPU or GPU across x86-64 and ARM64 hardware.