AI BriefingGemma 4 Transcription
Key point
An example of transcribing Gemma 4 E2B audio on macOS using MLX and mlx-vlm.
Details
A uv run example was introduced that runs the 10.28GB Gemma 4 E2B model on macOS with MLX and mlx-vlm to transcribe an audio file.
The execution example takes the form uv run --python 3.13 --with mlx_vlm --with torchvision --with gradio mlx_vlm.generate, feeding file.wav into the google/gemma-4-e2b-it model with the prompt Transcribe this audio. Options such as --max-tokens 500 and --temperature 1.0 are also used together.
In a 14-second WAV demo, the model transcribed a short voice memo fairly plausibly, but it misheard some expressions. It's also noted that the original phrases "This right here..." and "... how well that works" were each interpreted as different words.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Underdog Releases 'Husky', a Model-Specific Inference Engine Up to 4.5x Faster than MLX