← Back to live feed1 story
A developer repurposed the EmbeddingGemma 2 multimodal model into a speech-to-text tool that can transcribe one hour of audio in approximately 13 seconds on a base M3 MacBook Air. The 740M parameter model was released by Google under an Apache 2.0 license and supports local Mac applications like DigUp and Foresight.
Sign in to suggest
Key sources
- SOURCE@tslshahir“transcribes an hour of audio in about 13 seconds: 288× real time”x.com
- SOURCE@googledeepmind“outperforming some specialist models more than twice its size”x.com
- SUPPORT@ivanfioravanti“Google Embedding Gemma 2 is the DeepSeek moment of embedding models”x.com
- SOURCE@alex_tra_memory“10,080 posts sorted into topics live, ~620/s on a MacBook”x.com
- SUPPORT@seraandroid“leverage Apple’s Neural Engine and this project working with EmbeddingGemma 2”x.com
- SUPPORT@fdaudens“A dev built DigUp, which is basically what search on your Mac should be”x.com
- SUPPORT@clementdelangue“Introducing EmbeddingGemma 2, our new open embeddings model for on-device use cases!”x.com
- SUPPORT@testingcatalog“uses local models, Gemma 4 E4B or Gemma 4 12B, to transcribe voice notes”x.com