Accelerating the search for antimicrobial peptides with Google Antigravity
deepmind.google·2h ago
AI Engineering
Hacker News·2d ago·5 min read
An hour of audio transcribed in about 13s on MacBook Air
TL;DR
EmbeddingGemma 2, a multimodal embedding model by Google, has been adapted into a speech-to-text model that transcribes an hour of audio in just 13 seconds on a MacBook Air, entirely on-device.
✦ Why It Matters
Consider implementing on-device transcription solutions to improve user experience in your applications today.
Key Takeaways
- Utilize EmbeddingGemma 2 for efficient audio transcription.
- Experiment with the model's capabilities on a MacBook Air.
- Consider training the reader model for improved accuracy.
- Integrate the transcription feature into your applications easily.
How It Works
The model retains its original audio encoder while a small reader processes short audio slices, mapping them to the nearest English words. This allows for rapid transcription, with an additional model handling punctuation and capitalization.