General-purpose speech recognition and multilingual transcription model
Whisper is an open-source neural net trained on 680,000 hours of multilingual and multitask supervised data collected from the web. It delivers robust automatic speech recognition (ASR) and language translation.
Pros
Industry benchmark for transcription accuracy and accents
Runs 100% offline with zero cloud audio leakage
Wide community tooling ecosystem (whisper.cpp, MacWhisper)
Cons
Larger models require 4GB+ VRAM for real-time speeds