Available Models
Weesper Neon Flow includes seven local recognition profiles. The download size is a useful planning measure; actual memory use also depends on your operating system and workload.
| Profile | Detail Level | Speed | Download Size | Language Coverage | Best For |
|---|---|---|---|---|---|
| Minimal | Everyday | Very fast | ~75 MB | Multilingual | Older hardware and quick notes |
| Compact | Everyday | Very fast | ~148 MB | Multilingual | Lightweight daily dictation |
| Compact+ (EN) | English-optimised | Very fast | ~145 MB | English only | Sharper everyday English transcription |
| Everyday | Detailed | Fast | ~190 MB | Multilingual | Fast everyday writing |
| Standard | Detailed | Fast | ~512 MB | Multilingual | Most users (recommended default) |
| Precision | High detail | Balanced | ~547 MB | Multilingual transcription | Polished notes, accents, and technical terms in the spoken language |
| Professional | High detail | Steady | ~3.0 GB | Multilingual | Accuracy-first professional work |
Minimal (EN), Extended, and Expert are legacy profiles retained for existing installations. They are not offered for new downloads. Compact+ (EN) is the current English-only profile; it is not suitable when you dictate in another language. Precision transcribes the spoken language and does not translate to English.
How to Change Models
- Open Settings → Recognition Quality
- Select your preferred profile
- The model loads automatically for your next dictation
Recommendations
- Apple Silicon Mac (M1+): Standard is the recommended default; try Precision or Professional for accuracy-first work
- Intel Mac / Windows: Standard or Everyday is usually the best balance
- Older hardware (4 GB RAM): Use Compact or Minimal
- Professional / medical / legal: Use Precision or Professional if your machine can handle the delay
Understanding Transcription Speed
When you release the hotkey, there is a short delay before your text appears. This is completely normal — Weesper Neon Flow is transcribing your audio locally on your device, and the time it takes depends on three factors:
| Factor | Impact on Speed | Impact on Accuracy |
|---|---|---|
| Recognition profile (Minimal → Professional) | Larger profiles are slower | Larger profiles are more accurate |
| Accuracy level | Higher accuracy takes longer | Better transcription quality |
| Your hardware (CPU, RAM) | Faster hardware = faster results | No impact |
The Speed–Quality Tradeoff
Think of it as a triangle: model power, accuracy, and hardware. You control the first two — adjust them to match your hardware:
- Powerful machine (Apple Silicon, modern CPU, 16 GB+ RAM) → Use Standard, Precision, or Professional with a higher quality setting
- Mid-range machine (8 GB RAM) → Standard + Everyday quality is the sweet spot
- Older hardware (4 GB RAM, older CPU) → Compact or Minimal + a faster quality setting keeps dictation responsive
Finding Your Sweet Spot
If transcription feels too slow after you release the hotkey:
- Go to Settings → Recognition Quality
- Try a lighter profile first — switch from Professional or Precision to Standard, Everyday, Compact, or Minimal
- Then adjust accuracy — lower it one step and test again
- Repeat until you find the right balance between speed and quality
The best configuration is the one where you barely notice the delay while still getting accurate transcriptions. Test with a few short sentences after each change.
Shorten Long Pauses
Enable Shorten long pauses in Settings → Recognition Quality to make long dictations faster by trimming silence before transcription.
Related
- Improving Accuracy — Tips beyond model selection
- Custom Dictionary — Add domain-specific vocabulary
- System Requirements — Check RAM and hardware recommendations