openai/whisper: Robust Speech Recognition via Large-Scale Weak Supervision

“It also requires the command-line tool ffmpeg to be installed on your system, which is available from most package managers” (Setup (README))

“base 74 M base.en base ~1 GB ~7x” (Available models and languages table)

“Whisper is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio” (README, first paragraph)

“large 1550 M N/A large ~10 GB 1x” (Available models and languages table)

“large 1550 M N/A large ~10 GB 1x” (Available models and languages table)

“large 1550 M N/A large ~10 GB 1x” (Available models and languages table)

“medium 769 M medium.en medium ~5 GB ~2x” (Available models and languages table)

“Whisper's code and model weights are released under the MIT License.” (License section)

“Whisper's performance varies widely depending on the language.” (Available models and languages section)

“You can download and install (or update to) the latest release of Whisper with the following command: pip install -U openai-whisper” (README, Setup)

“the codebase is expected to be compatible with Python 3.8-3.11 and recent PyTorch versions” (Setup section)

“on Windows using Chocolatey (https://chocolatey.org/) choco install ffmpeg” (Setup section)

“You may need rust installed as well, in case tiktoken does not provide a pre-built wheel for your platform.” (Setup section)

“The following command will transcribe speech in audio files, using the turbo model:” (Command-line usage)

“small 244 M small.en small ~2 GB ~4x” (Available models and languages table)

“small 244 M small.en small ~2 GB ~4x” (Available models and languages table)

“tiny 39 M tiny.en tiny ~1 GB ~10x” (Available models and languages table)

“tiny 39 M tiny.en tiny ~1 GB ~10x” (Available models and languages table)

“tiny 39 M tiny.en tiny ~1 GB ~10x base 74 M” (Available models and languages table)

“However, the turbo model is not trained for translation tasks. If you need to translate non-English speech into English, use one of the multilingual models” (README, Command-line usage)

“the turbo model is an optimized version of large-v3 that offers faster transcription speed with a minimal degradation in accuracy” (Model table, paragraph after it)

“medium 769 M medium.en medium ~5 GB ~2x large 1550 M N/A large ~10 GB 1x” (Available models and languages table)

“medium 769 M medium.en medium ~5 GB ~2x large 1550 M N/A large ~10 GB 1x” (Available models and languages table)

“base 74 M base.en base ~1 GB ~7x small 244 M small.en small ~2 GB ~4x” (Available models and languages table)