EmbeddingGemma 2 Turned Into On-Device Speech-to-Text Model

Developers turned Google's multimodal EmbeddingGemma 2 into a speech-to-text model that processes an hour of audio in about 13 seconds on a MacBook Air, with roughly 5% word error rate and fully on-device processing.

• Zero-training trick lifts EmbeddingGemma-2 to MAEB top2026-10-28
• Google launches local-first AI meeting note-taker Foresight2026-10-08
• Google Releases Open Multimodal Embedding Model EmbeddingGemma 22026-10-06
• Google opens Gemini and Search to third-party apps2026-10-28
• Google AI Mode gives evasive answers on 2020 election2026-10-28