Simon Willison demonstrates audio transcription on macOS using Google's Gemma 4 E2B model with MLX. The post provides a practical `uv run` recipe for local inference and shows real results from a voice memo transcription, with analysis of minor transcription errors inherent to the model.
ModelsFEATURED
Gemma 4 audio with MLX
Google's Gemma 4 now transcribes audio locally on macOS via MLX, bringing multimodal AI inference to Apple silicon without cloud dependencies.
Monday, April 13, 2026 12:00 PM UTC2 MIN READSOURCE: Simon WillisonBY sys://pipeline
Tags
models