Configuration

Configure Digital Human behavior, voice traits, scripted responses, DTMF, and IVR simulation

Digital Humans are highly configurable. You can change their accents, languages, background noises, emotions, speaking speeds, and more. This page covers every configuration lever available to you.

Voice & Speech Settings

These settings control how the Digital Human sounds during a conversation with your agent.

Setting	Description	Example Values
Language	The language the Digital Human speaks	English, Spanish, Mandarin, Hindi, French
Accent	Regional accent applied to speech	British, Southern US, Latin American, Australian
Speaking Speed	How fast or slow the Digital Human talks	Slow, Normal, Fast
Emotion	The emotional tone of the conversation	Calm, Frustrated, Anxious, Cheerful, Angry
Volume	How loudly the Digital Human speaks	Quiet, Normal, Loud
Background Noise	Ambient sounds during the conversation	Office, Street, Airport, Car, None

Combining settings creates realistic personas. A frustrated customer calling from a noisy airport behaves very differently from a calm customer in a quiet home office — and your agent should handle both.

Intent & Success Criteria

Every Digital Human needs a clear intent and success criteria to produce meaningful test results. Some generation paths (for example workflow-based scenarios) may store empty description or success criteria; see the Create digital human and Generate digital humans API references for how your payload is validated.

Defining Intent

The intent is what the Digital Human wants to accomplish in the conversation. Write it as a clear scenario description:

Intent: "I'm calling because I was charged twice for my monthly subscription.
I want a refund for the duplicate charge and confirmation that it won't
happen again next month."

Defining Success Criteria

Success criteria are the conditions that determine whether the test passed. They should be specific and measurable:

Success Criteria:
- The agent acknowledged the duplicate charge
- The agent initiated a refund or escalated to billing
- The agent provided a confirmation number or next steps

The more specific your success criteria, the more actionable your simulation results will be. Avoid vague criteria like “the agent was helpful” — instead, define exactly what “helpful” means for that scenario.

Advanced Behaviors

Digital Humans can do far more than just talk. These advanced behaviors let you test scenarios that standard conversational testing can’t reach.

Scripted Responses

Configure a Digital Human to provide specific responses depending on what your agent says. This is useful when you need deterministic behavior for regression testing.

1. Define trigger phrases

   Specify what the agent might say that should trigger a scripted response — for example, “Can I have your account number?”

2. Map scripted replies

   Define the exact response the Digital Human should give when the trigger is matched — for example, “Sure, it’s 4829-3371.”

3. Set fallback behavior

   Decide what happens when no trigger matches. The Digital Human can continue with natural conversation or stay on script.

DTMF Codes

Digital Humans can send DTMF (touch-tone) codes during a call. This is critical for testing agents that require keypad input — such as entering an account number, selecting a menu option, or confirming a PIN.

DTMF Sequence: "1" → "4829" → "#"
Trigger: After the agent says "Please enter your account number followed by the pound sign"

Silence Simulation

Configure a Digital Human to stay silent for a specified duration. This tests how your agent handles dead air — does it re-prompt the customer? Does it escalate? Does it hang up too early?

Separately, the API stores allow_silence_tool (boolean) and silence_tool_instructions (string) on each digital human. When allow_silence_tool is true, the voice runtime may use a silence tool according to its own rules. Use the literal string "default" for instructions to mean “built-in product behavior”; any other non-empty string is custom guidance for that runtime. Ending a call is analogous but not identical: allow_end_call_tool plus optional hangup_instructions (often null when you do not want custom hangup copy). Whether the silence tool actually runs is enforced in the simulation execution layer, not in the API middleware alone.

IVR System Simulation

Digital Humans can simulate an IVR (Interactive Voice Response) system so your agent can navigate through it. This is the inverse of the typical setup — instead of a human calling your agent, your agent is calling into a phone tree, and the Digital Human plays the role of that phone tree.

1. Define the IVR menu tree

   Map out the menu structure: “Press 1 for Sales, Press 2 for Support, Press 3 for Billing.”

2. Configure agent navigation

   Set the expected path your agent should take through the IVR to reach the correct department.

3. Validate the outcome

   Define success criteria for the agent successfully navigating the IVR and reaching the intended endpoint.

Configuration Combinations

The power of Digital Humans comes from combining these settings. Here are a few example configurations:

Frustrated bilingual caller

Language: Spanish → English mid-call
Emotion: Frustrated
Speed: Fast
Intent: Billing dispute, wants to speak to a manager
Success: Agent de-escalates and resolves without transfer

Silent customer on hold

Silence: 15 seconds after greeting
Emotion: Neutral
Intent: Waiting for agent to re-engage
Success: Agent re-prompts within 10 seconds, doesn’t disconnect

DTMF account verification

DTMF: Sends account number when prompted
Emotion: Calm
Background: Office
Intent: Check account balance
Success: Agent correctly reads back the balance

IVR navigation test

Mode: IVR simulation
Menu: 3-level phone tree
Intent: Agent must reach “Billing → Refunds → Existing Case”
Success: Agent arrives at correct endpoint within 60 seconds

Next Steps

Use Cases

See real-world patterns for building effective Digital Human test suites.

Generate via API

Create Digital Humans programmatically with the generation endpoint.