OpenAI Whisper Large v3 Turbo (2024年)（ウィスパーラージv3ターボ）

概要

OpenAI Whisper Large v3 Turbo は2024年10月OpenAI発表・Industry-leading 809M parameters 2024年OpenAI Whisper Large v3 Turbo MIT speech recognition position確立。Industry-leading 8x faster vs Large v3 + 99 languages + 4-decoder-layer pruned (vs Large v3 32) + Greek+Thai+Cantonese+Vietnamese improved + MIT license。

主な特徴・仕組み

• Industry-leading OpenAI (2015 SF California) 2024年10月 release (Whisper-v3-turbo)
• Industry-leading 809M parameters signature (vs Large v3 1.55B)
• Industry-leading 8x faster vs Large v3 signature
• Industry-leading 99 languages support signature (slightly less than Large v3 100)
• Industry-leading 4-decoder-layer pruned signature (vs Large v3 32)
• Industry-leading minimal quality degradation vs Large v3
• Industry-leading 16kHz audio input + 30-second chunks
• Industry-leading transcription + translation to English signature
• Industry-leading MIT license signature
• Industry-leading HuggingFace native + OpenAI API
• Industry-leading Whisper Large v3 → Whisper Large v3 Turbo evolution

まとめ

OpenAI Whisper Large v3 Turbo = 2024年10月OpenAI発表のpremium MIT speech recognition model。Industry-leading 809M parameters vs Large v3 1.55B + 8x faster vs Large v3 + 99 languages support + 4-decoder-layer pruned vs Large v3 32 + minimal quality degradation + 16kHz audio + 30-second chunks + transcription + translation to English + MIT license + HuggingFace native + OpenAI API + Whisper Large v3 → Whisper Large v3 Turbo evolution + OpenAI 2015 SF Sam Altman 9 years position確立。