---
title: "Fine-tune narration with the Pronunciation Dictionary"
slug: "/help/demos/edit/guides/pronunciation-dictionary"
topic: "demos"
last_updated: "2026-09-08T09:54:03.250Z"
---

## About This Feature

The **Pronunciation Dictionary** lets your workspace teach Walnut's text-to-speech (TTS) narrator how to say tricky words — product names, acronyms, foreign terms, proper nouns — so guide narration and presenter clips come out right every time. Rules are applied automatically on the server before speech is synthesized, so every existing and future TTS clip benefits without re-recording.

### Snapshot 📸

**Feature Status:** Included with Text-to-Speech narration

**Role Access:** Admin, Editor

**Location:** Any TTS panel in the guide/annotation editor → **Pronunciation Dictionary** (in the panel header)

---

## What a Rule Contains

Every rule maps one **written form** to how it should sound. You can build a rule in either of two ways:

- **Phonetic spelling** — respell the word the way it sounds (for example, `Walnut → wall-nut`).
- **Record yourself saying it** — click **Say it out loud** on the record row and read the word into your mic. Walnut transcribes the recording and generates up to three phoneme candidates you can preview and pick from.

Each rule also has:

- **Priority** — controls which rule wins when two rules could match the same run of text (see *Overlapping matches* below).
- **Preview voice** — pick any narrator voice to hear the rule before you save.

---

## Quick Start

### Open the dictionary

1. In the editor, open any annotation's **Text-to-Speech** panel.
2. Click **Pronunciation Dictionary** in the panel header.

### Add a rule

1. Click **Add rule**.
2. Enter the written form (the word or phrase that appears in your guide text).
3. Enter its pronunciation — either type a phonetic spelling or click **Say it out loud** to record it.
4. If you recorded, review the three phoneme candidates. Each has a red recording indicator on the row while capture is live and its own error copy if capture fails; you can retry any candidate.
5. Pick a **preview voice** and click **Preview** to hear the rule.
6. Click **Save**. Changes stage on the Save button — you can add or delete several rules and commit them together.

### Edit or delete a rule

Open the dictionary, hover a row, and use the row's edit or delete controls. Deletions also stage on **Save** so you can undo before committing.

Validation errors appear directly under the offending row.

---

## Overlapping Matches

Two rules can sometimes both match a run of text (for example, one rule for `AI` and another for `AI Mode`). Each rule has a **priority** — the highest-priority matching rule wins. Reorder rules or edit their priority when a longer, more specific rule is being masked by a shorter one.

---

## FAQs

**Do rules apply retroactively to guides I already recorded?**
Yes — rules are applied on the server every time TTS is synthesized. Any playback of an existing guide picks up your latest dictionary.

**Are rules per-workspace or per-demo?**
Per-workspace. A rule you add applies to every demo, playlist, and Deal Room in your workspace.

**Which languages are supported?**
Rules are applied per-voice at synthesis time, so any voice that supports phoneme hints will honor the rule. Custom recordings work in any language your microphone can capture.

**Can I hear the rule before saving?**
Yes — pick a preview voice on the rule and click **Preview**. Walnut renders the rule inside a short sample so you can compare before committing.