Dictation Speed & Latency
TalkWriter is designed to deliver polished text within seconds of you finishing speaking. This page explains what happens during that time and how to minimize delays.
Latency Breakdown by Pipeline Stage
When you release the Fn key (or toggle off in Live mode), your dictation passes through several stages before text appears. Here is how long each stage typically takes:
| Stage | What Happens | Typical Latency | Depends On |
|---|---|---|---|
| Audio capture | Mic records and streams audio | ~0ms (real-time) | Microphone quality |
| Network upload | Audio sent to cloud servers | 50-150ms | Internet speed, distance to server |
| Speech-to-Text | Soniox converts audio to raw text | 100-300ms | Dictation length, language |
| AI Polish | AI cleans grammar, removes fillers | 300ms-1.5s | Text length, TalkTone enabled |
| Network download | Polished text sent back to your Mac | 50-100ms | Internet speed |
| Paste | Text inserted at cursor | ~10ms | Target app responsiveness |
Total Typical Latency
| Dictation Length | Without TalkTone | With TalkTone (Pro) |
|---|---|---|
| Short (1-2 sentences) | 0.5-1 second | 1-2 seconds |
| Medium (1 paragraph) | 1-2 seconds | 2-3 seconds |
| Long (multiple paragraphs) | 2-4 seconds | 3-5 seconds |
What Affects Speed
Internet Connection
Your internet speed is the most significant factor in dictation latency.
| Connection | Expected Impact |
|---|---|
| Fiber / fast broadband | Minimal latency, best experience |
| Standard broadband | Good performance |
| Mobile hotspot (4G/5G) | Slightly higher latency, still usable |
| Satellite internet | Noticeable delay (500ms+ added) |
| Slow or unstable Wi-Fi | Intermittent delays or failures |
Dictation Length
Longer dictations require more processing time for both speech-to-text and AI Polish. Short dictations (a sentence or two) are nearly instant.
AI Polish and TalkTone
- AI Polish on: Adds 300ms-1.5s for text cleanup.
- AI Polish off: Skips cleanup, fastest possible output (but raw, unformatted text).
- TalkTone (Pro): Adds additional processing for style adaptation.
Server Load
During peak usage times, processing may take slightly longer. This is rare and typically adds less than 500ms.
How to Optimize for Speed
| Optimization | Impact | How |
|---|---|---|
| Use a fast internet connection | ⭐⭐⭐⭐⭐ | Connect via Ethernet or strong Wi-Fi |
| Keep dictations short | ⭐⭐⭐⭐ | Dictate 1-3 sentences at a time |
| Disable TalkTone | ⭐⭐⭐ | Settings > TalkTone > Off (keeps AI Polish) |
| Disable AI Polish | ⭐⭐ | Settings > AI Polish > Off (raw output, fastest) |
| Close bandwidth-heavy apps | ⭐⭐ | Pause large downloads or video streaming |
Frequently Asked Questions
Why is my first dictation of the day slower? The first dictation may have slightly higher latency as connections are established. Subsequent dictations are faster.
Can I see processing time for each dictation? TalkWriter does not currently display per-dictation timing. The pill overlay indicates processing progress visually.
Is there a way to get instant results? Turn off AI Polish in Settings for the fastest possible output. You will receive raw speech-to-text results with no cleanup.
Does dictation speed vary by language? Slightly. High-resource languages (English, Spanish, Mandarin) process fastest. Less common languages may take marginally longer.
Was this helpful? Let us know at support@talkwriter.ai