//
Transcribe audio with OpenAI's Whisper
Whisper is OpenAI's speech recognition model that rivals professional transcription services. Run it locally for free, private transcription of any audio or video.
Install OpenAI's Whisper package.
pip install openai-whisper
# For faster inference, also install:
pip install faster-whisperOne command to transcribe any audio file.
whisper audio.mp3 --model large-v3
# With faster-whisper:
faster-whisper audio.mp3 --model large-v3đĄ First run downloads the model (~3GB for large-v3).
Larger models are more accurate but slower.
# tiny: 39M params, ~32x realtime, lower accuracy
# base: 74M params, ~16x realtime
# small: 244M params, ~6x realtime
# medium: 769M params, ~2x realtime
# large: 1550M params, ~1x realtime, best accuracy
whisper audio.mp3 --model mediumâ Transcription is slow
â Use faster-whisper which is 4x faster. Or try a smaller model size.
â Poor accuracy
â Use large-v3 model. Ensure clean audio input.