Correct your names and terms. Keep the words you meant.

One sound, two meanings

Suppose your speech model writes “cloud” when you say “Claude”. Adding the name alone is not enough: you still need to talk about cloud storage.

The dictionary can associate the recognized form “cloud” with both meanings: the assistant name, spelled “Claude”, and the ordinary computing meaning, spelled “cloud”. The decision model reads the context and selects among the meanings you have supplied.

This example explains the intended behaviour; it is not a guarantee for every sentence. The model can choose incorrectly, and a missing meaning limits the choices it can make.

Build a dictionary from your dictation

1. Start with a speech model. Set one up in Models and dictate a few examples of the names and technical terms you use. History keeps your recordings and transcription attempts.
2. Choose a dictionary model. On the Dictionary page, choose the cloud language model used to generate suggestions. Configure its provider access in Settings. Supported OpenAI access also includes ChatGPT sign-in, subject to your plan.
3. Get suggestions and review them. “Get suggestions” uses that speech model’s raw history to propose additions, updates or removals. Edit the proposals, dismiss unwanted ones and apply the rest. Nothing is applied automatically.
4. Check the meanings. Include the correct spelling, a useful definition and the forms your speech model actually produces. Where an ordinary word competes with a name, include that literal meaning too. You can also use “Add an entry” to create this knowledge by hand.
5. Enable contextual corrections. Choose a decision model in Settings → Corrections & formatting and turn on dictionary correction. Dictionary generation and live correction are separate jobs, with separate model choices.

For a larger starting set, “Learn from audio” can transcribe recordings you select and propose dictionary entries. Check the personal dictionary guide for imports, review and recovery.

Test corrections and words that should stay

Try a sentence using the name, then one using its ordinary competing word. Include phrases you actually dictate, not just the name in isolation. Inspect the result in History before relying on it for important text.

• The wrong spelling stays: check that the recognized form is associated with the intended meaning for the selected speech model.
• An ordinary word changes: check its competing literal meaning and definition. Pinning a name does not give it priority over the other meanings.
• Results change after switching models: learned associations are specific to the speech model. Pinned knowledge is shared across models.
• Correction fails: Entune skips dictionary changes and shows a notice. Other enabled processing steps can still apply their changes. The original speech transcript is saved before processing begins.

Advanced, explicitly approved direct mappings can bypass context. Use the direct-mapping reference before enabling that exception. A pinned or single-meaning entry is not automatically a direct mapping.

To judge a setup, count both wanted corrections and unwanted changes. Entune’s published comparison and methodology reports those separately; it does not measure overall speech-recognition accuracy.

What “learns your words” means

Entune builds a dictionary you can inspect and edit. It does not retrain your speech model. A decision model selects stored spellings for matching words; it does not invent a replacement that is missing from the dictionary.

Generating suggestions sends source transcripts and relevant dictionary entries to your chosen cloud language model. Cloud decision models receive the text needed for enabled processing, even if speech recognition is local. Laya is a separate local decision engine. See local dictation setup and the full data and privacy reference.

There is no Entune account. Cloud providers may still require an account, API key, subscription or usage payment.

Try it with your own dictation

Install Entune, set up a speech model, allow the permissions it requests, and choose a shortcut. Hold the shortcut to record; release it to transcribe and paste.

Puts Entune in Applications and opens it. The first start can take up to a minute.

Full installation steps and pip alternative