MacBook Air
AssessPlatforms
Apple laptop used as the reference hardware for the local transcription benchmark.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by research breakthroughs, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 7Hacker News·10/13/2026breakthroughEmbeddingGemma 2 turned into an on-device speech-to-text model
A team adapted Google's EmbeddingGemma 2 multimodal embedding model into a speech-to-text system that can transcribe an hour of audio in about 13 seconds on a base MacBook Air. The approach keeps the original audio encoder and trains only a small reader, with transcription and punctuation handled fully on device.