TOOLORA / Local reading edition
faster-whisper
A CTranslate2-based Whisper implementation with CPU/GPU quantization, batched transcription, word timestamps and voice-activity filtering.
Overview
faster-whisper runs compatible Whisper models through CTranslate2 to produce text and timed segments. Its Python interface supports batched transcription, word timestamps, Silero VAD filtering and language detection, with quantized CPU or GPU computation. It suits self-managed services and offline batch processing. Model files must be prepared, and GPU acceleration requires matching CUDA, cuDNN and CTranslate2 versions; speed depends on the model and hardware.
Official sources and references
- Official websitehttps://github.com/SYSTRAN/faster-whisper
Descriptions summarize public information; consult official websites for features, pricing and regional availability.