Gemma 4 runs local audio transcription on Mac via MLX

Google’s 10.28 GB Gemma 4 E2B model can now transcribe audio files locally on macOS using MLX and mlx-vlm. Simon Willison shared a single uv run command that handles the entire setup and execution without manual dependency management.

A test on a 14-second voice memo produced mostly accurate results, though it stumbled on a couple of phrases - mishearing “this right here” as “this front here.” The errors were phonetically plausible, suggesting the model handles casual speech reasonably well for its size but still has rough edges on ambiguous audio.