# Character Definition

NavTalk supports the customization of digital personas' personalities, language styles, welcome messages, automatic response rules, and contextual memory capabilities through the `instructions` field in the `session.update` request, enabling highly tailored behaviors of digital entities.

By configuring the `instructions`, you can set the following:

* The behavioral style and professional background of the digital persona
* The response language (Chinese, English, Japanese, etc.)
* The initial welcome message
* The function call trigger rules
* Contextual memory and personalized preferences
* Voice emotional characteristics

## Configuration Entry: `session.update` Request Body

When establishing a connection with the OpenAI Realtime API, send the session with configured instructions:

```javascript
const instructions = `
  ${activeCharacter.characterDefinition}.
  Please respond in ${userLanguage}.
  Please greet the user with: "${activeCharacter.greeting}".
  When encountering requests beyond conversational capabilities, 
  automatically trigger function calls for:
  - Multimedia content generation (images/audio/video)
  - Real-time data queries (weather/stocks, etc.)
  - Business system interactions (CRM/ERP, etc.)

  Current context: ${messageConfig}
`;

const sessionConfig = {
  type: "session.update",
  session: {
    instructions,
    voice: "nova", // Optional voice type (see the voice list below)
    temperature: 1,
    modalities: ["text", "audio"],
    input_audio_format: "pcm16",
    output_audio_format: "pcm16",
    input_audio_transcription: {
      model: "whisper-1"
    }
  }
};

socket.send(JSON.stringify(sessionConfig));

```

## Parameter Descriptions:

| parameter name        | type   | example                                                                              | explanation                                                                      |
| --------------------- | ------ | ------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------- |
| `characterDefinition` | string | You are a gentle and rational psychologist                                           | This is my input: Used to set personality, tone, professional background, etc    |
| `userLanguage`        | string | zh-CN / en-US / ja-JP                                                                | Control the output language                                                      |
| `greeting`            | string | "Hello, I am your intelligent assistant."                                            | An optional welcome message that supports variables                              |
| `function call rules` | string | The natural language description of the timing for invocation.                       | Describe the scenarios in which functions are called automatically               |
| `messageConfig`       | object | Contains a JSON object with a summary of historical dialogues, user preferences, etc | Used to inject the current contextual information (supporting AI + user content) |
| `voice`               | string | "nova"                                                                               | Specify the desired voice style (see below)                                      |

### **Behavioral Style and Professional Background of the Digital Persona**

 By configuring `characterDefinition`, you can define the digital persona's personality traits, tone style, knowledge background, and communication habits.

| Example setting                                                                                                           | Description                                                                                 |
| ------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
| You are a gentle and rational counselor, skilled in listening and empathy.                                                | This is my input: For mental health applications                                            |
| You are a witty and humorous virtual host, skilled at creating a lively atmosphere.                                       | This is suitable for live interaction and entertainment content.                            |
| You are a business consultant with 20 years of experience, known for your clear, concise, and professional communication. | The improved English version is: "For enterprise customer service and knowledge inquiries." |

### **Response Language Setting(**`userLanguage` **)**

You can specify the dialogue language for the digital persona using the `userLanguage` parameter, which supports various languages such as Chinese, English, Japanese, etc.

| language | parameter value |
| -------- | --------------- |
| Chinese  | `zh-CN`         |
| English  | `en-US`         |
| Japanese | `ja-JP`         |

{% hint style="info" %}
**Note:**

* After setting, all response content will be generated in that language.
*  If this setting is absent in the `instructions`, the system will default to automatically recognizing the user's input language.
  {% endhint %}

### **Initial Welcome Message (`greeting`)**

The digital persona can initiate the first conversation proactively by setting a welcome message after establishing a connection.

| parameter name | type   | example                                                         |
| -------------- | ------ | --------------------------------------------------------------- |
| `greeting`     | string | "Hello, I am your intelligent assistant. How can I assist you?" |

### **Function Call Trigger Rules**

NavTalk supports seamless integration between the digital persona and external systems using OpenAI Function Calling technology (e.g., databases, weather APIs, order inquiries).

> For details, see the Function Calling section: [Tool Call](https://navtalk.gitbook.io/api/real-time-digital-human-api/images-and-media)

### **Contextual Memory Strategy (`messageConfig` vs. `conversation.item.create`)**

NavTalk supports contextual memory capabilities, enabling the digital persona to understand the context of previous conversations.

> For details, see the Maintaining Historical Conversations section: [Maintaining Historical Dialogue](https://navtalk.gitbook.io/api/real-time-digital-human-api/images-and-media-1)

### **Supported Voice Styles (`voice`)**

NavTalk offers various high-quality voice synthesis styles. You can freely choose the digital persona's voice type using the `voice` parameter.

| Voice Name | Description of Style                                             | Sample listening link                                        |
| ---------- | ---------------------------------------------------------------- | ------------------------------------------------------------ |
| `alloy`    | Neutral authority, suitable for business/popular science         | [🔊](https://cdn.openai.com/API/voice-previews/alloy.flac)   |
| `shimmer`  | Bright and energetic female voice, suitable for customer service | [🔊](https://cdn.openai.com/API/voice-previews/shimmer.flac) |
| `ballad`   | Warm narrative male voice                                        | [🔊](https://cdn.openai.com/API/voice-previews/ballad.flac)  |
| `coral`    | Professional and Clear Broadcast Voice                           | [🔊](https://cdn.openai.com/API/voice-previews/coral.flac)   |
| `echo`     | A rendition with echoes                                          | [🔊](https://cdn.openai.com/API/voice-previews/echo.flac)    |
| `ash`      | The stable and trustworthy male voice                            | [🔊](https://cdn.openai.com/API/voice-previews/ash.flac)     |
| `sage`     | Intelligent and gentle female voice                              | [🔊](https://cdn.openai.com/API/voice-previews/sage.flac)    |
| `verse`    | Rich in rhythm, suitable for reciting poetry                     | [🔊](https://cdn.openai.com/API/voice-previews/verse.flac)   |