whisper-ctranslate2

Why we included this project

If your team already drives OpenAI's Whisper through its command line, this client is the low-effort way to get much faster transcriptions without relearning anything: it accepts the same arguments, so switching mostly means pointing your scripts at a different binary. Under the hood it runs the CTranslate2 and faster-whisper engines, which reach about four times the original's speed at similar accuracy while using less memory, and batching input files widens that gap considerably. It also adds a VAD filter that skips silence, speaker diarization so transcripts show who said what, and live transcription from a microphone, handy for meetings. A prebuilt Docker image comes with the small, medium, and large-v2 models, and you can load your own fine-tuned or quantized Whisper checkpoints, which matters if your team has trained a specialized variant.