faster-whisper

faster-whisper is a reimplementation of OpenAI's Whisper that transcribes audio several times faster while using less memory. Developers embed it to add quick, private, offline transcription.

Faster-whisper is a reimplementation of OpenAI's Whisper speech-to-text model built on CTranslate2, a fast inference engine for Transformer models. It produces the same transcriptions as the original Whisper but runs up to about four times faster using less memory, and 8-bit quantization can speed it up further on both CPU and GPU. It is a Python developer library rather than a finished app: you call it from your own code to turn audio into timestamped text. It supports all Whisper model sizes, word-level timestamps, batched transcription for higher throughput, built-in Silero voice-activity detection to skip silence, and automatic language detection. Models download automatically from the Hugging Face Hub in CTranslate2 format. It is widely used as the transcription back end for many other open-source projects, including real-time streaming tools, diarization pipelines, and OpenAI-compatible transcription servers.