AI Models
Model Settings
Configure shared transcription cleanup, model loading, and enhancement timeouts.
Open AI Models, click the gear button, then select Transcription or Enhancement. These shared settings are separate from a Mode's model, language, prompt, and output choices.
Transcription
| Setting | Behavior |
|---|---|
| Whisper Prompt | Language-specific recognition hints for local Whisper models. Choose a language before editing; this is not an AI enhancement prompt. |
| Remove Filler Words | Edit the words removed before formatting and enhancement. Remove every entry to stop this cleanup. |
| Add Space After Paste | Add a trailing space to pasted output. |
| Voice Activity Detection (VAD) | Help local transcription separate speech from silence. It does not configure every cloud provider's silence handling. |
| Prewarm model (Experimental) | Prepare the selected local model at launch or wake to reduce loading delay. |
| Timeout | How long batch cloud transcription can wait. Options range from 10 seconds to 30 minutes; the default is 30 seconds. This does not set local or streaming timeouts. |
See Filler Words for list management.
Enhancement
| Setting | Behavior |
|---|---|
| Skip short transcriptions | Skip enhancement of short live dictations. Enabled by default. |
| Minimum words | Threshold from 1 to 15, default 3. Live dictation with that many words or fewer skips enhancement. Respond output bypasses this shortcut. |
| Timeout duration | Maximum wait for an enhancement attempt, from 3 to 60 seconds; default 7 seconds. |
| On timeout | Choose Fail immediately or Retry. Retry allows up to three attempts in total and applies to timeouts, not every provider error. |
When ordinary enhancement fails or times out, VoiceInk can retain the original transcription with a warning. Imported-file processing has its own workflow and does not apply the live dictation short-text threshold. Local CLI has a separate command timeout.
See Mode Settings for per-Mode choices.