Offline models
| Message | What to do |
|---|---|
| The model for this language is not downloaded yet… | Press Download in the strip under the languages, or open the chooser and pick a model that is already on the computer. |
| Download failed: … | Check the internet connection and press the download button again. |
| Could not start the local model. Try downloading it again. | The model files may be incomplete. Download the model again. |
| Whisper runs on the processor — very slow. | Without an NVIDIA card Whisper takes a long time per phrase. Use Parakeet v3, GigaAM v3, T-one or Vosk for dictation, or a smaller Whisper size. |
| CUDA is not installed — Whisper runs on the processor. | Press "Download CUDA" in Whisper's details in the chooser. |
Sizes and download places are on offline models.
Soniox
"Soniox is not available now — dictating through Google. Next time Soniox is tried again." means Soniox could not be used for this recording, usually because of the connection, so Google took over without losing your words. If it happens every time, check the internet. Without any connection neither online engine works; switch to LOCAL.
When the month's credits are used up, dictation goes through Google until they renew; see Soniox credits.
Free plan limits
- "Today's free limit is used up — 2,000 words." Dictation is available again after midnight, or right away with Pro.
- "The free version transcribes one file and one link a day." The FILES allowance also resets at midnight.
- "The free version takes recordings up to 60 minutes." Cut the recording into parts, or use Pro.
- "Several files or links at once is part of Pro." On Free, transcribe them one at a time.
The full comparison is on Free and Pro.
FILES
- "No model downloaded yet." Press "Choose model…" on FILES and download one; it only needs downloading once.
- "Model path saved but the file is missing: …" The model file was moved or deleted. Choose or download a model again.
- A file ends with ERROR. Open SETUP → Files and press "Test engines". If the faster engines fail, choose "Plain processor · CPU", which is built in. See file transcription speed.