Start with a short acknowledgement

1. Acknowledge before the main answer

A brief “mm” or “hmm” can let the user hear that the character has responded before the main answer arrives. Use a closed-mouth sound rather than a short phrase such as “Sure” or “Wow.” Keep content that needs precise lip movement in the main reply. Set the opening, then enable its fast playback: the character instructions put the acknowledgement on its own first line, and pipeline_config.tag_audio_enabled lets an eligible short acknowledgement play before the main answer. The switch does not generate the opening text.

2. Update Chat Style and Hard Rules

Add word choice and variation to Chat Style. Add the first-line format, real newline, and tool-only exception to Hard Rules. Merge the following into the existing avatars[].instructions, retaining the character’s identity and normal response rules. See Write spoken instructions for the complete structure.

## Chat Style — Short acknowledgement
For every reply that contains speech, put one closed-mouth acknowledgement on the first line, then start the substantive reply on the next line. For English replies, use mm, hmm, or mhm. For other languages, use a short closed-mouth acknowledgement appropriate to that language. Keep the opening natural for the reply language. Do not substitute open-mouth words or phrases such as Sure, Wow, or No problem.
Choose a different acknowledgement from the most recent spoken assistant reply. Compare the whole word ignoring punctuation and English letter case. Tool-only turns do not count. If no previous spoken opening is available, choose freely.

## Hard Rules — Opening and spoken output
The first line contains only the acknowledgement, optionally followed by punctuation. Put a real newline immediately after it, with no blank line, label, tag, quotation marks, or other text. Do not output the literal characters backslash and n.
Apply the character's answer-first, apology, empathy, and length rules to the body after the first line. The opening does not count toward the body's sentence limit. Add it once per spoken reply. Omit it when the response contains only tool calls and no speech.
Output only spoken words. Do not add stage directions, Markdown, or JSON. Keep the body in the user's requested language.

Start with a small set you can listen to: “mm,” “hmm,” and “mhm” for English, or “嗯” and “嗯嗯” for Chinese. Match the opening to the reply language. For other languages, choose and test an appropriate closed-mouth sound rather than adding arbitrary short words. Replace conflicting instructions such as “keep the whole reply on one line,” “start with any three words,” or “always produce speech.” Update their examples too. Appending a new rule while leaving the old demonstration intact creates a conflict.

3. Enable the fast opening audio

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

Merge pipeline_config into the Create Session request. tag_audio_enabled belongs inside that object and must be a boolean. It defaults to false. Preserve your other settings and test in a new Session.

4. Check the text, then listen

The generated reply should contain a real line break:

hmm
The speaker is a better fit for your desk.

Inside a JSON string, \n encodes a newline. After parsing, it must be a line break, not the literal characters backslash and n. In a plain text editor, enter an ordinary newline. The fallback opening limits are two Han characters and ten English letters; platform configuration may change these limits. The “嗯” and “hmm” examples on this page fit those fallback limits. Mixed text must satisfy both limits. Punctuation does not count. Only the first complete speech segment is considered. The service does not extract “hmm” from a longer opening sentence. Once the text is correct, listen for a clear acknowledgement and a smooth transition into the answer.

5. Try it with Playground

Use Avatar Playground to prepare the character and voice and start a conversation. Merge the prompt into Avatar Instructions and enable pipeline_config.tag_audio_enabled in the project’s Create Session request. End the current experience, then start again to test. Try a practical question, a correction, an emotional remark, and a language change. Check the first line, variation from the previous spoken reply, and the character of the answer. If your application has a silent tool-only turn, test that it remains silent.

Overview

A1: Streaming Avatar

W1: Streaming World

Blazing Fast Video Generation

Start with a short acknowledgement

1. Acknowledge before the main answer

2. Update Chat Style and Hard Rules

3. Enable the fast opening audio

4. Check the text, then listen

5. Try it with Playground

​1. Acknowledge before the main answer

​2. Update Chat Style and Hard Rules

​3. Enable the fast opening audio

​4. Check the text, then listen

​5. Try it with Playground

1. Acknowledge before the main answer

2. Update Chat Style and Hard Rules

3. Enable the fast opening audio

4. Check the text, then listen

5. Try it with Playground