Recommended AI Models
A practical starting point for choosing fast transcription and enhancement models in VoiceInk.
Transcription
Models for turning speech into text
| Model | Best use |
|---|---|
| Parakeet model | Best first choice for fast, real-time offline transcription. |
| Whisper large v3 turbo | Strong Whisper option when you want broad accuracy and reliable results. |
Local vs cloud transcription
Local models are best when privacy and low latency matter. Cloud models can help with less common languages or when local processing is not suitable, but they depend on your internet connection.
Use a local model when privacy, speed, and offline use are the priority. This is the recommended path for daily dictation on Apple Silicon Macs.
Use a cloud model when local processing is too slow, the language is less common, or your Mac is not a good fit for local transcription.
Enhancement
Models for improving the final text
| Provider | Models |
|---|---|
| Groq | gpt-oss-120b |
| Cerebras | gpt-oss-120b, GLM-4.7 |
| Gemini | gemini-3.1-flash-lite |
| OpenRouter | gpt-oss-120b |
Keep enhancement fast
If enhancement regularly takes longer than two seconds, switch providers. Fast inference keeps dictation feeling natural.
Choose a fast transcription model
Start with Parakeet model for local transcription.
Pick a low-latency enhancement provider
Start with Groq and gpt-oss-120b when you want fast cleanup after transcription.
Check the full session timing
If dictation feels slow, open transcript history, select transcripts, and use the analyze option in the bottom toolbar.
Free tiers are enough to start
The listed providers have free tiers that are useful for testing and daily use before you decide whether paid models are worth it.
Developer note
You can understand what is taking time during a VoiceInk session by opening transcript history, selecting transcripts, and using the analyze option in the bottom toolbar.