Vosk
Real-time transcription on a CPU. Runs on a Raspberry Pi.
95
sovereigntyVosk is a lightweight offline speech-recognition toolkit built for constrained environments — it runs in real time on a CPU, works on Android, iOS and single-board computers, and its models are tens of megabytes rather than gigabytes. It will not match Whisper on accuracy for difficult audio. It is the right answer when you need speech recognition on a device, offline, with no GPU and no network, which is a requirement no managed API can satisfy at any price. Apache-2.0.
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST
LicenseApache-2.0
PricingFree, Apache-2.0.
Open sourceYes
Self-hostableYes
Local-first dataYes
What it does well
- +Real-time on CPU — no GPU required at all
- +Tiny models; runs on phones and single-board computers
- +Fully offline, which some products require by law
- +Around twenty languages supported
Where it falls short
- −Noticeably less accurate than Whisper on hard audio
- −Smaller language coverage
- −No built-in diarization
Vosk as an alternative to
Where Vosk shows up in our comparisons, and how it ranked.
Vosk head-to-head
Straight comparisons against the tools people weigh it against.