faster-whisper
A reimplementation of OpenAI's Whisper model on the CTranslate2 inference engine, running several times faster with lower memory use than the reference code. It is a common engine inside self-hosted transcription services.
faster-whisper is a reimplementation of OpenAI’s Whisper model built on the CTranslate2 inference engine. It produces the same transcripts as the reference Whisper code but runs several times faster and uses less memory, which is why it has become a common engine inside self-hosted and on-prem transcription stacks.
Because it changes only the inference layer, not the model weights, its word error rate matches the underlying Whisper model — the gain is speed and efficiency, not accuracy. It supplies the raw decoding step in larger pipelines such as WhisperX, which add word-level timestamps and speaker diarization on top.