Skip to main content

Additional features: Voice

When to use​

Enable Voice when you want hands-free conversations, spoken summaries, or quick “talk & respond” flows—e.g., field ops, multilingual help desks, kiosks, or accessibility-first experiences.

What it does​

  • Adds a Use voice mode button in the chat composer.
  • Captures your speech, transcribes it for the agent, and speaks the agent’s reply back.
  • Optional voice summaries after each response.
  • Optional announcements when the agent uses tools (“I’m creating a ticket…”).
  • Supports interruptions/barge-in so you can talk over the agent to clarify or stop.

Enable in the UI​

  1. Open your agent → Edit.
  2. Under Additional features, click Add.
  3. Toggle Voice configurations ON.
  4. A gear icon appears next to “Voice configurations” for detailed setup.

Configure voice (gear icon)​

  • Enable voice as summary — read a short audio summary after each agent message.
  • Announce Tool Usage — the agent speaks short status lines before/after tool calls.
  • Language — sets the language for both transcription and synthesis (e.g., Hindi).
  • Transcriber config — choose the Transcriber provider (e.g., Azure).
  • Synthesizer config — choose the Synthesizer provider and voice name (e.g., Azure / hi-IN-SwaraNeural).
  • Conversation config
    • Enable Agent Interruptions — allow barge-in while the agent is speaking.
    • Interrupt speech duration (s) — minimum voice activity to trigger an interruption (e.g., 0.5).

Tip: Pick the same language for transcription & TTS for best accuracy and naturalness.

Using voice in chat​

  1. Click Use voice mode in the composer to start a voice session.
  2. Speak normally; you’ll see live activity in the bottom voice dock.
  3. The agent replies with audio + text.
  4. If interruptions are enabled, start speaking to cut in; or use the dock controls to mute/end.

Best practices​

  • Short turns win: 1–3 sentences per spoken reply; offer to send a longer text/Artifact if needed.
  • Confirm write actions: “Should I submit this ticket now—yes or no?”
  • Domain hints: Tell the agent to announce critical tool usage (searching KB, creating ticket, etc.).
  • Accessibility: Pair with Follow-up Suggestions for quick tap responses during or after voice.
  • Testing: Verify the selected voice name and language sound correct for your audience.

Troubleshooting​

  • No mic input / button disabled: Give the browser microphone permission; check OS input device.
  • No audio playback: Verify output device; some browsers require a user gesture (click) before audio.
  • Wrong language/transcript quality: Set Language correctly and use a matching voice.
  • Speech cuts off / can’t interrupt: Enable Agent Interruptions and increase the interrupt duration slightly (e.g., 0.7–1.0s).
  • Long tool operations: Enable Announce Tool Usage so users hear progress; consider higher tool timeouts in Tools → gear.