> ## Documentation Index
> Fetch the complete documentation index at: https://docs.anyreach.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Pronunciations and text normalization

> Control how the agent pronounces words and reads numbers and prices.

Two settings on the text-to-speech (TTS) stack control how the agent speaks: `pronunciations` overrides how specific words are said, and `text_normalization` rewrites prices in the LLM's output before they reach the TTS engine. Both live on the agent version's `stt_llm_tts` stack alongside the rest of your [voice and model config](/agents/voice-and-model-config).

## Pronunciations

`Tts.pronunciations` is a dictionary that maps a word to the pronunciation the TTS engine should use for it. Use it to fix brand names, acronyms, and other terms that the default voice mispronounces.

| Field            | Type                        | Default | Description                                          |
| ---------------- | --------------------------- | ------- | ---------------------------------------------------- |
| `pronunciations` | `object` (string -> string) | `null`  | Maps each source word to its pronunciation override. |

```json theme={null}
{
  "tts": {
    "models": [ /* ... */ ],
    "pronunciations": {
      "Anyreach": "any reach",
      "SQL": "sequel"
    }
  }
}
```

When `pronunciations` is omitted or `null`, no overrides are applied.

## Text normalization

`TextNormalizationConfig` rewrites the LLM's text before it is spoken so that prices are read aloud naturally. Normalization runs in a separate text-normalization-service that processes the text before it reaches the TTS engine.

It is **disabled by default** (`enabled` is `false`). Set `enabled` to `true` to turn it on.

| Field                   | Type                         | Default  | Description                                                    |
| ----------------------- | ---------------------------- | -------- | -------------------------------------------------------------- |
| `enabled`               | `boolean`                    | `false`  | Enables text normalization for TTS.                            |
| `price_verbosity`       | `formal` \| `conversational` | `formal` | How prices are spoken.                                         |
| `shorthand_cents_extra` | `array<int>`                 | `[]`     | Extra cents endings that use shorthand under `conversational`. |

```json theme={null}
{
  "text_normalization": {
    "enabled": true,
    "price_verbosity": "conversational",
    "shorthand_cents_extra": [25, 75]
  }
}
```

### Price verbosity

`price_verbosity` controls how monetary amounts are spoken. It applies only to the `$`, `£`, and `€` symbols.

| Value            | Behavior                                                                                                                       |
| ---------------- | ------------------------------------------------------------------------------------------------------------------------------ |
| `formal`         | Full long form, for example `$9.99` becomes "nine dollars and ninety-nine cents".                                              |
| `conversational` | Retail shorthand for `$1`–`$99` amounts whose cents fall in the shorthand set, for example `$9.99` becomes "nine ninety-nine". |

<Note>
  Under `conversational`, only amounts in the `$1`–`$99` range with a cents value in the shorthand set are spoken as shorthand. Other amounts fall back to the long form.
</Note>

### Shorthand cents endings

The shorthand set used by `conversational` verbosity defaults to `{49, 50, 95, 99}`. Add more cents endings with `shorthand_cents_extra`. Each value must be an integer in the `0`–`99` range. `shorthand_cents_extra` is ignored when `price_verbosity` is `formal`.

| Limit                       | Value                  |
| --------------------------- | ---------------------- |
| Default shorthand cents set | `49`, `50`, `95`, `99` |
| Allowed value range         | `0`–`99`               |

Values are validated, de-duplicated, and sorted. Out-of-range values or non-integers are rejected.

<Info>
  The legacy flat `enable_text_normalization` flag is auto-migrated. When an older agent version is loaded, that boolean is lifted into the nested `text_normalization` config as `{"enabled": <value>}`, so you only need to set the nested form going forward.
</Info>

## Related

<CardGroup cols={1}>
  <Card title="Voice and model config" icon="waveform-lines" href="/agents/voice-and-model-config">
    Configure the STT, LLM, and TTS stack that these settings extend.
  </Card>
</CardGroup>