Raw voice transcription captures exactly what you said. Every "um," every false start, every run-on sentence. Accurate, but messy.
Enhance sits between your transcription and your clipboard. After Vext converts your speech to text, Enhance runs it through a local AI model that cleans up the noise — filler words, broken sentence structure, redundant phrasing — without changing what you meant.
What Enhance does
The processing pipeline:
- You speak into Vext
- The transcription engine converts speech to text
- Enhance sends that text to a local LLM with a cleanup instruction
- The processed result is what gets pasted
Before Enhance:
"Okay so basically I think we should um probably refactor the the authentication middleware because right now it's like doing too many things and uh it's hard to test"
After Enhance:
"We should refactor the authentication middleware. It's doing too many things and it's hard to test."
Same meaning. Same tone. Shorter, cleaner, and ready to send.
What Enhance does NOT do
Enhance is not a rewriter. It does not:
- Change your meaning or intent
- Add information you did not say
- Convert casual tone to formal language
- Restructure your argument
- Remove technical terms or jargon
It removes noise. The raw transcription is always saved alongside the enhanced version, so you can compare or revert.
When to use Enhance
Enable for:
- Slack messages and emails — removes verbal clutter, reads as written text
- AI prompts — filler words do not help LLMs, clean prompts work just as well
- Documentation and notes — clean text is easier to reference later
- Code review comments — professional tone without extra effort
Disable for:
- Creative writing where raw speech patterns matter
- Verbatim transcription needs (legal, medical, interviews)
- When you want to review the exact words before sending
Enhance + translation
When both Enhance and a target language are enabled, cleanup and translation happen in a single pass. You speak messy Russian, and clean English text appears at your cursor.
This is more than sequential processing — the AI model handles both tasks together, producing more natural translations than a clean-then-translate pipeline would.
How it runs locally
Enhance is powered by local LLMs that run entirely on your Mac's Apple Silicon GPU:
| Model | Size | Notes |
|---|---|---|
| Gemma 3 4B | 2.8 GB | Default, best quality |
| Qwen 3 4B | 3.2 GB | Strong multilingual support |
| LLaMA 3.2 3B | 2.4 GB | Strong general-purpose |
| Gemma 3 1B | 0.8 GB | Fastest, lightweight |
| Phi-3.5 Mini | 2.8 GB | Compact, strong reasoning |
No internet connection required. No data leaves your Mac. The model processes your text in milliseconds on Apple Silicon.
You can also use an OpenAI-compatible API if you prefer cloud processing — but local is the default.
Meeting summaries
For meeting recordings, Enhance powers the summarization feature. After a meeting ends, it extracts:
- Key points — the main decisions and information discussed
- Action items — tasks assigned, with owners when mentioned
Both the full raw transcript and the summary are saved. Summaries are generated from the complete transcript, so they capture context from the entire meeting.
Tips for getting the most out of Enhance
Leave it on by default. The cleanup is conservative enough that it rarely changes meaning. Start with it enabled everywhere and only disable for specific cases where you need verbatim output.
Speak naturally. Enhance works better with natural speech than with careful, deliberate enunciation. The more natural you speak, the more noise there is for Enhance to clean — and the more dramatic the improvement.
Do not over-enunciate to help Enhance. It does not need help. Speak the way you would talk to a coworker. The AI model understands conversational language.
Check the raw transcript occasionally. Especially early on, compare the enhanced and raw versions to build confidence in what Enhance changes and what it preserves.
Try it
Enhance is included with Vext — no separate purchase or add-on. The free trial includes 100 dictations with full Enhance support. Download Vext and try it with your next Slack message.