VoiceInk is slow: find which stage takes the time, then fix that one

By · Published

A VoiceInk dictation passes through several stages, and any of them can be the slow one: loading the model, transcribing, cleaning the text, AI enhancement, and pasting. Open Dashboard, View Insights, Model Performance to see which. If transcription is slow, use Parakeet V3 or a smaller Whisper model. If the final text is slow and the raw text is quick, the delay is AI enhancement.

I built YapToText, a dictation app for the Mac, so read what I say about VoiceInk with that in mind. Every fact about VoiceInk here comes from its own documentation at tryvoiceink.com, read on October 11, 2026.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

Find the slow stage

  • Dashboard, View Insights, Model Performance. Compares models over today, the last 7 or 30 days, this year or all time.
  • History, Analyze. Timing for a single recording.

Raw transcription fast but final output slow means AI enhancement. Everything slow from the first word means the transcription model. The VoiceInk history page covers the History view.

If transcription is slow

  • Use Parakeet. VoiceInk's docs name Parakeet V3 or Parakeet V2 for fast local dictation.
  • Try a smaller Whisper model if a large one is slow.
  • Prewarm the model. An experimental setting in AI Models, gear, Transcription that prepares the model at launch or wake.
  • Keep Voice Activity Detection on for local models, to separate speech from silence.
  • On an Intel Mac. Local transcription is usually fastest on Apple silicon. The docs say cloud transcription providers are often much faster on Intel.
  • First use. A large local model can take longer the first time it is prepared or loaded.

The best VoiceInk model page and the local models page go through the choices.

If AI enhancement is slow

Enhancement is a second model call after transcription. To shorten it:

  • Turn it off for simple dictation Modes.
  • Skip short transcriptions. A setting in AI Models, gear, Enhancement.
  • Adjust Timeout Duration to what you will tolerate.
  • Watch the retry setting. On timeout, Retry can make up to three attempts, which adds to the wait.
  • Change provider or model to a faster one.

If you use Local CLI as the enhancement provider, it has its own timeout, 45 seconds by default.

If a file is slow

Long audio and video files take longer than live dictation. The Transcribe view shows each file's state, Loading model, Processing audio, Transcribing, Enhancing, Completed or Failed, so you can see where it is waiting. The shared transcription Timeout applies to batch cloud requests, not to local processing. The VoiceInk file transcription page covers the queue.

If cloud transcription is slow

Then the speed depends on the provider's latency, the length of the audio and your connection, by VoiceInk's own description. A local model removes the connection from the equation on Apple silicon. The cloud providers page lists what VoiceInk supports.

How YapToText shows the same thing

YapToText, the free app I built, runs transcription and cleanup on an Apple silicon Mac, so there is no provider latency. Each History entry shows its stop-to-text time, split into transcription, AI cleanup and delivery, which is the same question VoiceInk's insights answer. I publish no speed numbers, because your Mac decides them.

Two settings matter for speed. Models stay loaded for two minutes after use by default, so a second dictation soon after the first skips the load. And Raw Transcription skips the cleanup model entirely. YapToText does not run on Intel Macs. VoiceInk not working covers failures that are not about speed.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

The short version

  • First. Dashboard, View Insights, Model Performance.
  • Slow transcription. Parakeet V3, a smaller Whisper, or prewarm.
  • Slow final text. AI enhancement. Skip short transcriptions, or a faster provider.
  • Intel Mac. A cloud transcription provider.
  • Files. Watch the queue state in Transcribe.

If VoiceInk changes these settings, open an issue on GitHub and I will update this page. Every one gets read.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

About the author. Ryleigh Newman is a developer whose hands make typing difficult and who dictates every day with YapToText, the app this site describes. Facts about other apps come from their own sites, with the date they were read. More about the project.

Sources

Questions

Why is VoiceInk so slow?

One stage is slow: loading the model, transcribing, AI enhancement or pasting. Dashboard, View Insights, Model Performance shows which.

What is the fastest VoiceInk model?

VoiceInk's docs name Parakeet V3 or Parakeet V2 for fast local dictation, and a smaller Whisper model if a large one is slow.

How do I speed up VoiceInk AI enhancement?

Turn on Skip short transcriptions, pick a faster provider or model, or disable enhancement for simple dictation Modes.

Is VoiceInk slow on Intel Macs?

Local transcription is usually fastest on Apple silicon. VoiceInk's docs say cloud transcription providers are often much faster on an Intel Mac.