Embedding Gemma Transcribe: 1 hour of audio in ~13 s (reported)
Why it matters
An independent project turned Google's EmbeddingGemma 2 into speech-to-text. They report 288x real time on a base M3 MacBook Air, fully on device, with ~5% word error rate vs 2-3% for leading models.
Read the source: eg2.haku.sh (opens in a new tab)
All hits from 13 Oct 2026More VoiceGet the daily hits by email