# What we measured.

These pages publish the measurements behind Earlyn's design decisions, with the method, the caveats and the data, so you can check the numbers we quote.

## [Turkish speech-to-text on a Mac](https://www.earlyn.app/research/turkish-meeting-transcription)

On 60 Turkish clips from Google's FLEURS test set, Whisper large-v3-turbo transcribed with an 8.2% word error rate against 10.9% for Qwen3-ASR 1.7B, both running locally on an Apple M2 Mac with 8 GB of memory.

## [How much disk screen memory takes](https://www.earlyn.app/research/screen-memory-storage)

At Earlyn's default 2,560-pixel capture, a busy hour of screen memory takes about 9 MB, which is around 160 stored moments at roughly 52 KB of picture and 4.5 KB of text each, and a still screen stores nothing.

## [Local search by meaning](https://www.earlyn.app/research/local-semantic-search)

In Earlyn's test of 12 search-by-meaning queries in English and Turkish, EmbeddingGemma 300M and bge-m3 ranked the right document first all 12 times, Apple's NLContextualEmbedding 4 times, and a converted multilingual-e5-small model once.

---
Canonical: https://www.earlyn.app/research · Updated 2026-10-02