Respond quickly with short vocal cues

A brief “mm” or “hmm” lets the avatar respond audibly before the main answer begins. It can make the pause feel more natural in a live conversation.

1. Add the cue to Speaking Style

In your Persona, add the following to Speaking Style. Keep the rest of the character’s personality and response habits.

## Speaking Style
For replies that include speech, begin with a short, closed-mouth acknowledgement such as mm, hmm, or mhm on a line of its own. Start the answer on the next line, with no blank line between them. Vary the sound rather than repeating the previous reply's opening.
Keep the first line to the sound alone, optionally followed by punctuation. Use a real line break. Do not add labels or stage directions. Keep the answer in the character's usual style and requested language. Omit the cue for a tool call that contains no speech.

Merge this into the existing section rather than adding a second Speaking Style section. Remove conflicting rules, such as “keep the entire reply on one line,” and update any examples that show the old format. Choose closed-mouth sounds. Words such as “Sure” or “Wow” need normal lip movement and belong in the main answer. For a character that speaks another language, choose and test a suitable closed-mouth sound for that language.

2. Enable fast playback

Set pipeline_config.tag_audio_enabled to true in your Create Session request:

{
  "pipeline_config": {
    "tag_audio_enabled": true,
    "tts_config": {
      "tts_voice_id": "longanhuan_v3.6"
    }
  }
}

Merge this fragment with your other settings. The flag is a boolean and defaults to false. Start a new session to apply the change. The Persona asks for the opening sound; the flag lets an eligible opening play quickly. Both are needed. The flag does not make the dialogue model generate the main answer faster.

3. Check the text, then listen

The generated reply should look like this:

hmm
That sounds like a good way to spend the afternoon.

Keep an English opening to no more than ten letters. Put the sound on a line of its own; “hmm, let me think about that” is a sentence, not a short vocal cue. In a JSON string, \n encodes a newline. After parsing, it should be a line break. In the Persona editor, press Enter to insert one. Listen for the sound followed by a smooth transition into the answer. Try a practical question, a correction, and an emotional remark. Check that the cue varies and that the answer still fits the character. If your application uses silent tool calls, check that those turns remain silent.

Overview

A1: Streaming Avatar

W1: Streaming World

Blazing Fast Video Generation

Respond quickly with short vocal cues

1. Add the cue to Speaking Style

2. Enable fast playback

3. Check the text, then listen