iohann.s.titov@gmail.com Ivan S. Titov OpenAI's Whisper speech-recognition model: end-to-end multilingual automatic-speech-recognition + translation, trained on 680k hours of weakly-supervised audio. The Python implementation ships transcribe() / detect_language() / load_model() APIs over a PyTorch backend. Imports as plain "whisper"; conflicts with dev-python/whisper (Graphite round-robin DB — same name, unrelated upstream). The Triton fastpath (triton>=2 from upstream's optional dep) is not pulled here; it would speed up long-form decoding but isn't packaged on this overlay yet. Functional fallback is pure-PyTorch. openai/whisper openai-whisper