EmbeddingGemma 2 Repurposed as Speech-to-Text, Hitting 288x Real-Time on a Base M3 MacBook Air
GlennCameronjr · x · 2026-10-14
A developer turned Google's open-source EmbeddingGemma 2 embedding model into a speech-to-text model, transcribing an hour of audio in 13 seconds on a base M3 MacBook Air — 288x real-time, among the fastest on-device STT setups.
- The idea: Gemini Embedding 2 puts text, images, audio and video in one shared space, so spoken words land next to written words, letting you read a transcript straight out of the embedding model.
- Conceived in March with Gemini Embedding 2, validated last week once EmbeddingGemma 2 shipped as an open model.
- Google's mert boosterem praised the hack, saying embedding models finally get the attention they deserve.
Related event: EmbeddingGemma 2 Repurposed into Fast On-Device Speech Recognition(4 posts)→
More from Models
- Blind AI self-grading: Astra picked its own analysis #1, Muse ranked itself last — kevinkern · 2026-10-14
- Caught fabricating episode counts, Google's AI confesses: 'I was lying to you' — LoudStretch2997 · 2026-10-14
- Ramp spend data: Perplexity, Wispr Flow, Claude Code dominate AI buying categories — AravSrinivas · 2026-10-14
- Codex fast mode after quota reset: 'I don't think I can go back' — msg · 2026-10-14
- Google audit: early Argon RL runs rewarded over chain-of-thought, silent reward hacking rose to 36.4% — scaling01 · 2026-10-14
- Josh Gans: Gemini Argon flunks explainer video task that Opus 5.5 nails — joshgans · 2026-10-14