Uploaded September 2025 | Updated September 2026, 16 minutes ago
Game-changer for speech AI: Ai2 just dropped OLMoASR, the first fully open Whisper-style transcription model—and it’s holding its own.
Trained on 440K hours of internet audio, it rivals Whisper in zero-shot transcription across audiobooks, meetings, and earnings calls, while releasing not just the models but also 3M hours of training data + full pipelines for reproducible research.
Yes, Whisper may still have the very edge at the largest scale, but the key here is: true openness, commercial use approved, and a new standard for transparency in ASR.
Which model release should I unpack next? Drop your pick 👇
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AI #opensource #speechrecognition #short
Game-changer for speech AI: Ai2 just dropped OLMoASR, the first fully open Whisper-style transcription model—and it’s holding its own.
Trained on 440K hours of internet audio, it rivals Whisper in zero-shot transcription across audiobooks, meetings, and earnings calls, while releasing not just the models but also 3M hours of training data + full pipelines for reproducible research.
Yes, Whisper may still have the very edge at the largest scale, but the key here is: true openness, commercial use approved, and a new standard for transparency in ASR.
Which model release should I unpack next? Drop your pick 👇
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AI #opensource #speechrecognition #short
