Softcatala/whisper-ctranslate2

Leverages CTranslate2's optimized inference engine with support for quantization (int8, float16) and batched processing to achieve 4-16x speedup over OpenAI's implementation while maintaining accuracy. Adds speaker diarization via pyannote.audio, Voice Activity Detection filtering, live microphone transcription, and custom fine-tuned model loading—all while maintaining CLI compatibility with the original Whisper client. Ships with Docker images pre-loaded with models and supports diverse hardware backends (Intel MKL, Apple Accelerate, NVIDIA GPUs, ARM64).

1,255 stars and 34,030 monthly downloads. Available on PyPI.