Whisper

Whisper

Whisper is OpenAI's open-weight automatic speech recognition (ASR) model trained on 680,000 hours of multilingual audio. It transcribes speech to text in 99 languages, translates non-English audio to English, and identifies the spoken language — all without per-language fine-tuning.

Key Features

• 99-language support — transcription and English translation from nearly any language
• Model sizes — tiny (39M) through large-v3 (1.5B) for speed/accuracy trade-offs
• Timestamps — word-level timestamps with the --word_timestamps flag
• Robust to noise — trained on diverse, in-the-wild audio including accents and background noise
• Local & private — runs fully on-device; no audio leaves your machine
• Ecosystem — faster-whisper, whisper.cpp, and WhisperX provide optimized runtimes

Quick Start

pip install openai-whisper

import whisper

model = whisper.load_model("base")  # tiny / base / small / medium / large
result = model.transcribe("audio.mp3")
print(result["text"])

# CLI usage
whisper audio.mp3 --model small --language English

Install via ai-supply

npx ai-supply add openai-whisper-speech-to-text

Curated mirror of the open-source Whisper project (MIT). Install upstream from the repository.