openai-whisper

What it is and what it does

Whisper is OpenAI's general-purpose speech recognition model that transcribes, translates, and identifies languages in audio. It uses a Transformer sequence-to-sequence architecture trained on diverse multilingual audio data, allowing a single model to handle multiple speech-processing tasks that traditionally required separate pipeline stages. The model comes in six sizes (tiny, base, small, medium, large, turbo) with English-only and multilingual variants, offering speed-accuracy tradeoffs from ~1 GB to ~10 GB VRAM.

You can use it via command-line (e.g., `whisper audio.mp3 --model turbo`) or Python API. It processes audio in 30-second sliding windows and supports language specification and translation tasks. Installation requires torch, numba, triton, and other heavy numerical libraries, plus ffmpeg on your system. The package is actively maintained and has no known vulnerabilities.

Use it for:

• Transcribe English audio files quickly using the turbo model for real-time or batch processing
• Translate non-English speech to English by specifying language and task parameters
• Identify the spoken language in an audio file before further processing
• Build a speech-to-text pipeline that handles multiple languages with a single model
• Process audio in Python with fine-grained control via lower-level APIs like detect_language() and decode()

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Whisper performs multilingual speech recognition, speech translation, and language identification using a Transformer model trained on diverse audio data.

Yes, if you have the system dependencies and can tolerate high install friction. Whisper is actively maintained, permissively licensed, widely used (top 5000 packages), and has no known vulnerabilities. The heavy numerical dependencies (torch, numba, triton) are unavoidable for the task. Install only if you need multilingual speech recognition or translation; for English-only transcription, smaller or specialized models may be more practical.

Install

openai-whisper on PyPI

pip install openai-whisper

uv add openai-whisper

poetry add openai-whisper

Installing openai-whisper

Before you install

High install friction: requires torch, numba, triton, and other heavy numerical dependencies. Also requires ffmpeg as a system dependency and may need Rust installed if tiktoken lacks a pre-built wheel for your platform. Maintenance is active with recent commits.

License in practice

MIT license permits commercial and private use with minimal restrictions, making it suitable for most projects.

Quickstart

pip install openai-whisper

import whisper
model = whisper.load_model("turbo")
result = model.transcribe("audio.mp3")
print(result["text"])

Requires ffmpeg installed on your system (apt/pacman/brew/choco/scoop). May also require Rust if tiktoken has no pre-built wheel for your platform.

Verify before relying

• Whether pre-built tiktoken wheels cover common platforms or if Rust compilation is frequently needed
• Real-world transcription speed and accuracy on languages beyond English
• Memory and compute requirements for models beyond the documented VRAM estimates

Package facts

Evidence: openai_whisper-20250625.tar.gz

Tags

More Artificial Intelligence packages

Further reading