Fact · Dictation in your apps
50megabytes per language model
Alpha Cephei says Vosk's per-language models are about 50 MB each, and that much bigger server models are also available.
Checked against its source on October 17, 2026
In the source's own words
“Portable per-language models are only 50Mb each, but there are much bigger server models available.”
Vosk Offline Speech Recognition API (Alpha Cephei), Alpha Cephei, 2026, Home page, introduction.
Open the original- What it measures
- Vosk is an offline recognizer that its home page says works on lightweight devices such as the Raspberry Pi. The size applies to the portable per-language models.
- What it does not show
- The page gives no word error rate, so this fact shows the download size and nothing about accuracy. Server models are larger.