Skip to main content
Vowen does almost nothing between recordings. It does not listen in the background or transcribe continuously. CPU use spikes while a dictation is transcribed and drops straight back. Memory use depends mostly on which speech model is loaded. A local model stays in memory between dictations so the next one starts instantly. Larger models use more memory. A cloud model uses none, because nothing runs on your machine. Speech models page, Local section, listing the English-only and multilingual local models

Transcription is slow

  1. The first run of a new model is slow. Vowen optimizes the model for your Mac the first time it loads. Parakeet Ultra takes 15 to 17 seconds. This happens once.
  2. Whisper Large v3 is slow by nature. Try Large v3 Turbo, or a Parakeet or Nemotron model.
  3. Parakeet and Nemotron are the fast local models on a Mac. Vowen keeps them loaded, so there is no start-up delay per dictation.
  4. Check Activity Monitor for other apps using the CPU. Heavy load can also stop macOS delivering shortcuts to Vowen. See common issues.

Cloud models

  • Check your connection, and check that Settings › Developer › Prefer IPv4 Connections is on.
  • Try another provider. Providers have busy periods.
  • Long recordings take longer to upload and transcribe.

AI enhancement

If the delay comes after transcription, it is the AI enhancement step, not the speech model. See AI issues.

High memory use

The speech model stays loaded between dictations on purpose. Unloading it would make every dictation wait for it to load again. To use less memory:
  1. Pick a smaller model. This makes the biggest difference. Whisper Base uses a fraction of what Large v3 uses.
  2. Use Parakeet or Nemotron. Both are about 0.6B parameters, much lighter than large Whisper models. Cohere Transcribe (about 2B parameters) is heavier.
  3. Turn on Settings › General › Resource efficient mode. The model loads only when you dictate, so each dictation takes a few extra seconds.
  4. Use a cloud model.
Resource efficient mode only appears where it helps: with a local Whisper model, or with Parakeet on Windows.

Laptop runs warm

Warmth during transcription is normal and stops when transcription finishes. If your machine is warm while Vowen is idle:
  1. Check Activity Monitor or Task Manager to see whether Vowen is using CPU at all.
  2. Check whether a meeting note is still recording. The menu bar or tray icon shows a red dot while one is running.
  3. Restart Vowen to clear a stuck helper process.
  4. If it continues, report it on Discord with your model settings.

Battery

  • Cloud models move the work off your machine.
  • Parakeet and Nemotron use less power than large Whisper models on a Mac.
  • Smaller models use less energy per dictation.
  • Resource efficient mode keeps less in memory, at the cost of slower dictations.