AI Models

Dicta supports multiple AI transcription models. Each model has different strengths in terms of speed, accuracy, and language support.

Available Models

WhisperKit Models

Based on OpenAI’s Whisper, these models offer excellent accuracy across many languages.

Tiny (~75MB)

Fastest, lowest accuracy
Good for quick notes
99+ languages

Base (~150MB)

Fast, good accuracy
Balanced option
99+ languages

Small (~500MB)

Moderate speed, better accuracy
Great for most users
99+ languages

Large v3 Turbo (~800MB)

Slower, highest accuracy
Best for important content
99+ languages

Parakeet Models

NVIDIA’s speech recognition models, optimized for speed and accuracy.

Parakeet v2 (~600MB)

Very fast and accurate
English only
Best choice for English speakers

Parakeet v3 (~700MB)

Fast and accurate
25 European languages
Great for multilingual European users

Choosing a Model

Consider these factors:

Speed vs Accuracy: Smaller models are faster but less accurate. Larger models are more accurate but slower.

Language needs: If you only use English, Parakeet v2 offers the best speed/accuracy ratio. For multiple languages, WhisperKit models are more versatile.

Storage space: Models are downloaded and stored locally. Larger models require more disk space.

Hardware: More powerful Macs can run larger models more efficiently.

Managing Models

Open Dicta settings
Navigate to Models
Download your preferred model(s)
Select the active model

You can have multiple models downloaded and switch between them as needed. Delete unused models to free up disk space.

Model Performance

All models run 100% on-device. Performance depends on your Mac’s hardware:

Apple Silicon Macs: Excellent performance with all models
Intel Macs: Good performance with smaller models, larger models may be slower

Our Recommendations

For most users: WhisperKit Small or Parakeet v2 (English) For accuracy: WhisperKit Large v3 Turbo For speed: WhisperKit Tiny or Parakeet v2 For multiple languages: WhisperKit Small or Large v3 Turbo