Local Models
FloWords includes powerful AI models that run entirely on your Mac. No internet connection required, no data sent to external servers.
Why Local Models?
Section titled “Why Local Models?”- 100% Privacy - Your voice never leaves your device
- No Internet Required - Works offline, anywhere
- No Subscription - Use unlimited, no per-minute fees
- Fast Response - No network latency
Whisper Models
Section titled “Whisper Models”OpenAI’s Whisper is the primary transcription engine, optimized for Apple Silicon.
Available Models
Section titled “Available Models”| Model | Size | RAM | Best For |
|---|---|---|---|
| Tiny | 75 MB | ~1 GB | Quick notes, low-end Macs |
| Tiny.en | 75 MB | ~1 GB | English-only, faster |
| Base | 142 MB | ~1.5 GB | General use, balanced |
| Base.en | 142 MB | ~1.5 GB | English-only, good balance |
| Small | 466 MB | ~2 GB | Better accuracy |
| Small.en | 466 MB | ~2 GB | English-only, great accuracy |
| Medium | 1.5 GB | ~5 GB | Professional work |
| Medium.en | 1.5 GB | ~5 GB | English-only, excellent |
| Large-v3 | 3 GB | ~10 GB | Maximum accuracy |
Choosing the Right Model
Section titled “Choosing the Right Model”Recommended: Tiny or Base
- Fast transcription
- Lower memory usage
- Good for short dictation
- Works well on older Macs
Recommended: Base or Small
- Good balance of speed and accuracy
- Suitable for most users
- Works on any Apple Silicon Mac
- Handles multiple languages
Recommended: Medium or Large
- Best accuracy
- Better with accents and noise
- Ideal for important documents
- Requires 8GB+ RAM
Downloading Whisper Models
Section titled “Downloading Whisper Models”- Open FloWords Settings (⌘,)
- Go to Model tab
- Click Download next to your desired model
- Wait for download to complete
- Select the model as your active transcription engine
Parakeet Models
Section titled “Parakeet Models”NVIDIA’s Parakeet models (via FluidAudio) offer an alternative transcription engine.
Features
Section titled “Features”- Optimized for Real-Time - Low latency streaming transcription
- Efficient - Lower memory footprint
- English-Focused - Best accuracy for English content
When to Use Parakeet
Section titled “When to Use Parakeet”- Real-time dictation where speed matters
- English-only transcription needs
- Macs with limited RAM
- When you want an alternative to Whisper
Enabling Parakeet
Section titled “Enabling Parakeet”- Open FloWords Settings
- Go to Model tab
- Look for Parakeet in the model list
- Download if not already available
- Select as your active model
Model Performance by Mac
Section titled “Model Performance by Mac”Apple Silicon (Recommended)
Section titled “Apple Silicon (Recommended)”| Mac | Recommended Model | Performance |
|---|---|---|
| M1 (8GB) | Base, Small | Excellent |
| M1 Pro/Max | Medium | Great |
| M2/M3 | Small, Medium | Excellent |
| M2/M3 Pro/Max | Large-v3 | Excellent |
Intel Macs
Section titled “Intel Macs”| Mac | Recommended Model | Performance |
|---|---|---|
| i5 (8GB) | Tiny, Base | Good |
| i7 (16GB) | Base, Small | Good |
| i9 (32GB+) | Small, Medium | Moderate |
Model Configuration
Section titled “Model Configuration”Language Settings
Section titled “Language Settings”- Auto-Detect - Whisper automatically detects the language
- Specify Language - Set a specific language for faster, more accurate results
To set your language:
- Go to Settings > Model
- Find Language dropdown
- Select your preferred language
- Or choose Auto for detection
Quality vs Speed
Section titled “Quality vs Speed”Balance transcription quality against speed:
- Beam Size - Higher values = better accuracy, slower speed
- Best Of - Number of candidates to consider
- Temperature - Controls randomness in predictions
Memory Management
Section titled “Memory Management”FloWords monitors your system memory and warns you if you’re running low.
Low Memory Warnings
Section titled “Low Memory Warnings”If you see memory warnings:
- Switch to a smaller model - Tiny or Base use less RAM
- Close other applications - Free up system memory
- Restart FloWords - Releases cached memory
Memory Usage Tips
Section titled “Memory Usage Tips”- Close unused browser tabs before dictating
- Quit memory-heavy apps (video editors, VMs)
- Consider upgrading RAM if consistently limited
- Use
.enmodels for English-only work (slightly more efficient)
Troubleshooting
Section titled “Troubleshooting”Model Download Fails
Section titled “Model Download Fails”- Check your internet connection
- Ensure you have enough disk space
- Try downloading a smaller model first
- Restart FloWords and try again
Transcription is Slow
Section titled “Transcription is Slow”- Switch to a smaller model
- Close other applications
- Check Activity Monitor for memory pressure
- Ensure your Mac isn’t thermally throttled
Poor Accuracy
Section titled “Poor Accuracy”- Try a larger model (Small → Medium)
- Reduce background noise
- Speak more clearly and at a steady pace
- Add custom terms to your Dictionary
Model Won’t Load
Section titled “Model Won’t Load”- Restart FloWords
- Re-download the model
- Check console logs for errors
- Free up system memory
Next Steps
Section titled “Next Steps”- Configure AI Enhancement to polish your transcriptions
- Set up Cloud Providers as a backup option
- Add Custom Dictionary for specialized terminology