Speech Models 25 vs 100 Languages · GPU/CPU

Faster (25 languages) or Slower, More Precise (100 languages): which local model to use

Start with the Faster model whenever your meeting uses one of its 25 supported languages. Switch to the optional Slower, More Precise model when you want coverage across up to 100 languages.

Local Model Selection Decision Path
Speech Models 25 vs 100 Languages · GPU/CPU
01 · Language Check

In the 25 Languages?

The Faster model is the default local model covering 25 core European languages.

02 · Extended Set

Need All 100 Languages?

Download the optional Slower, More Precise model covering up to 100 languages.

03 · Hardware

GPU-First Execution

Automatic mode chooses a supported GPU first; CPU runs on the 100-language model run slower than real time.

04 · Verify

Inspect Names & Numbers

No benchmark is published - verify critical terms directly against your audio.

Stellar Recorder - Local Diarized Transcript & Executive Reader
Faster (25) · More Precise (100)
Executive Reader displaying local transcript output with named speakers and playback waveform
CriteriaFaster Model (25 Languages)Slower, More Precise Model (100 Languages)
Role in Stellar Recorder Default local speech modelOptional local download
Language coverage 25 languages100 languages (includes all 25)
Speed & precision Fast turnaround on CPU or GPUSlower, more precise across 100 languages
Device behavior Automatic picks supported GPU firstFull CPU runs can exceed recording length

All 25 languages in the Faster model

The Faster model covers Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish and Ukrainian. The Slower, More Precise model extends support to 100 languages total.

Hardware expectations & speed notes

Automatic device selection prioritizes a supported GPU first (including migrated older profiles without an explicit CPU lock). Running the full Slower, More Precise 100-language model purely on CPU can take longer than the meeting recording itself - choose the Faster 25-language model or a supported GPU whenever language permits.

How to compare accuracy on your own audio

Because Stellar Insights has not published an independent accuracy benchmark and microphone acoustics vary widely, run both models on a short consented sample of your own team calls and inspect domain jargon, names and numbers side by side.

Switch local speech models in Stellar Recorder

  1. Open Model Setup in Stellar Recorder settings.
  2. Select the Slower, More Precise (100-language) model as your optional local model (or keep the Faster 25-language default).
  3. Complete the download - setup validates your Microsoft Visual C++ runtime and GPU availability automatically.
FAQ

Frequently asked questions

Can I have both models installed?

Check Model Setup in Stellar Recorder to inspect which downloaded speech models are active on your machine.

Does the interface language change the speech model?

No. Switching among the 100 interface languages never changes your speech recognition model automatically - if a model lacks language support, the app asks you first.