Voice Channel in Copilot Studio | Copilot Studio Day 56

Day 56 of 365 Days of Copilot Studio explains the Voice Channel in Copilot Studio: how to build AI agents that can listen, understand, and respond by voice so callers can talk instead of type.

Chat moved support into text. Voice moves it back into conversation. A voice-enabled Copilot Studio agent answers calls, understands natural speech, accepts phone keypad input, retrieves business data, completes tasks, and hands off to a human when needed.

This lesson builds on Day 54: Custom Apps and Day 55: Omnichannel Deployment. Voice is one of the most powerful channels in that omnichannel plan.

[!NOTE] Voice is the conversation. Telephony is the connection. You need both for a complete voice experience.

Why Voice Channel?

Many users still prefer to call. They want to explain a problem, check an order, book an appointment, reset access, or reach support without navigating menus or typing long messages.

Reason	What it enables	Example
Talk instead of type	Callers speak naturally and hear a spoken answer.	Where is my order?
Faster resolution	The agent retrieves data and acts during the call.	Order 12345 was delivered on July 28, 2024.
Flexible input	Speech plus phone keypad covers every caller.	Say Support or press 2.
Connected support	AI handles routine calls, humans handle exceptions.	Escalate with full conversation context.

What Are We Building?

The target is a voice-enabled Copilot Studio agent: a caller dials in, the agent greets them, understands speech or keypad input, uses topics, knowledge, and actions, and responds by voice.

1. Build the agent: topics, knowledge, and actions for the top call reasons.
2. Enable voice: turn on the voice channel and choose Basic or Real-time Voice.
3. Configure voice: set voice, language, greetings, DTMF, silence, and interruption behavior.
4. Configure telephony: connect phone numbers through PSTN or Teams telephony.
5. Test with real calls: validate speech, keypad, silence, branches, and handoff.
6. Publish and operate: release the voice workstream and monitor calls.

How Voice Works

Every voice conversation follows the same loop: voice in, intelligence, voice out.

1. User speaks: Whats my order status?
2. Agent understands: converts speech to intent.
3. Topic, Knowledge, Action: matches a topic, reads approved content, calls an action or API.
4. Agent responds: Your order is on the way! spoken naturally.

Same Agent. Different Experience.

The same Copilot Studio agent can serve chat and voice. The knowledge, actions, and integrations stay shared. Only the experience changes.

Area	Chat	Voice
User input	User types: Create a support ticket for me.	User speaks: Create a support ticket for me.
Agent output	Sure! I will create a ticket for you.	Speaks: Sure! I will create a ticket for you.
Strength	Rich text, cards, links.	Hands-free, natural, accessible.

One agent. Multiple conversation experiences.

Choose the Right Voice Experience

Copilot Studio offers two voice modes. Choose for the scenario, not by habit.

Mode	Orchestration	Conversation style	Best for
Basic Voice	Classic orchestration	Structured conversations	IVR and predefined flows like Press 1 for Sales, Press 2 for Support.
Real-time Voice	Generative orchestration	Natural conversations	Open, low-latency dialogue like I need help with my order.

Structured or conversational - choose for the scenario.

Basic Voice: Structured and Predictable

Basic Voice is ideal for predictable call flows: defined flows, IVR menus, controlled responses, and topic-driven branches.

• Defined flows: Welcome! Press a number or speak your option.
• IVR menus: 1 Sales, 2 Support, 3 Billing.
• Controlled responses: the agent stays inside approved prompts.
• Topic-driven: each key maps to a tested topic.

Best for structured conversations.

Real-time Voice: More Natural Conversations

Real-time Voice is designed for natural, low-latency conversations with generative orchestration.

• Speak naturally: callers use everyday language.
• Fluid conversation: real-time, low-latency responses keep the dialogue moving.
• Generative orchestration: dynamic, intelligent turn-taking.
• Natural two-way conversation: I need to check my order status becomes Sure! Your order is on the way and will arrive tomorrow.

Let Users Simply Talk

Instead of navigating menus - 1. Track Order, 2. Account Info, 3. Billing, 4. Speak to Agent - the caller just asks: Where is my order?

1. Customer speaks: uses natural language.
2. AI understands: identifies intent.
3. Order request identified: check order status, retrieve order details, prepare response.

Turn everyday questions into real results.

What Is DTMF?

DTMF (Dual-Tone Multi-Frequency) lets callers interact using the phone keypad - press 1, 2, 3, and so on. Sometimes pressing a number is faster than speaking.

Key	Destination
1	Sales
2	Support
3	Billing

Give Callers a Choice

Combine speech and keypad so callers can use whichever is easier in the moment.

• Say Support: use natural language.
• Press 2: use the phone keypad.
• Same destination: same intelligent agent, great user experience.

Speech plus keypad equals flexible voice experiences.

From Traditional IVR to Conversational AI

Traditional IVR	Copilot Voice
Call, Listen, Press Number. Press 1 for Sales, Press 2 for Support, Press 3 for Billing, Press 0 for Operator.	Call, Speak Naturally, Get Help. I need help with my order becomes Sure! I can help with that.

Move beyond rigid phone menus.

Enable Voice in Copilot Studio

1. Agent: select your agent.
2. Settings: go to settings.
3. Voice: choose voice.
4. Enable Voice: turn it on, then pick Basic Voice or Real-time Voice.

• Basic Voice: classic orchestration, structured conversations, ideal for predictable scenarios.
• Real-time Voice: generative orchestration, natural conversations, ideal for dynamic open-ended scenarios.

Choose the voice experience your agent needs.

Voice Is Not Telephony

Capability	What you configure	Includes
Enable Voice	Voice authoring capabilities	Create voice topics, design conversations, use natural language.
Configure Telephony	Phone connectivity	Connect phone numbers, set up calling provider, configure inbound and outbound.
Fully Voice-Enabled Experience	Both together	People can call, have natural conversations, and create real business impact.

Handle Voice-Specific Situations

Voice calls need graceful recovery. Do not let the conversation reach a dead end.

Situation	Caller behavior	Agent response
Silence detected	Caller is quiet or no input is detected.	Are you still there? How can I help you?
Speech not recognized	Speech is unclear or not understood.	Sorry, I did not catch that. Could you please repeat it?
Unknown keypad input	Caller presses an invalid or unexpected key.	That number is not valid. Please try again or say what you need.

• Understand the situation.
• Respond helpfully.
• Keep the conversation going: I can help! Lets try that again.

What If the Caller Says Nothing?

1. Caller silent: no speech or input is detected.
2. No input detected: the system detects silence after a few seconds.
3. Agent prompts again: Are you still there? with a follow-up prompt.

• Keep the conversation going.
• Be helpful and patient.
• Provide clear next steps.

Design graceful recovery for silence.

Let Users Interrupt Naturally

Real conversations include interruptions. Support barge-in so callers do not have to wait.

1. Agent speaking: Let me help you with that.
2. Caller starts talking: Actually, I have a different question.
3. Agent stops and listens: pauses output and captures the new intent.
4. Conversation continues: Great! Lets get that sorted out.

More natural. Less robotic.

Shape the Voice Experience

Area	What to design	Options
Voice output	Choose the voice, tone and language your agent uses.	Select voice and language, set tone and style, customize greetings.
Conversation behavior	Control how your agent talks and responds.	Set response style, manage interruptions (barge-in), define turn-taking.
DTMF	Define how the agent handles keypad input.	Set up keypad options, map actions to numbers, provide clear prompts.
Silence handling	Manage what happens when the caller says nothing.	Set silence timeout, define retry prompts, choose escalation path.

Design how your agent sounds and behaves.

Speak to More Customers

One voice agent can serve multiple languages: English (Hello!), Spanish (Hola!), French (Bonjour!), and many more supported languages.

One agent. Multiple languages.

Voice Can Do More Than Talk

A voice agent is not just a responder. Connected to business systems, it takes action during the call.

Caller: Whats my order status? Voice agent: Let me check that for you. The agent connects to knowledge (SharePoint, Docs), actions (Power Automate), data (Dataverse, CRM), and APIs (external systems), retrieves Order 12345 - Delivered Jul 28, 2024, then speaks: Your order 12345 was delivered on July 28, 2024.

Powered by your business data. Turn conversations into business actions.

AI When Possible. Human When Needed.

1. Caller: I need help with my order. Calls and speaks naturally.
2. Copilot agent: Let me check that for you. Understands, helps and takes action.
3. Escalation: Let me connect you to a specialist. Transfers with full conversation context.
4. Human agent: Hi! I can see your order details here. Continues the conversation seamlessly.

Conversation context travels along: caller intent, key details such as order number and issue, previous steps, and full conversation history. No need to repeat yourself.

Escalate without starting the conversation over.

Voice Agent in Action

1. Customer calls: I need help with my order.
2. Check my order status: customer asks for their order status.
3. Agent retrieves order: checks CRM or ERP data. Let me check that for you.
4. Provides status: Order 12345 - Processed, Shipped, Arriving tomorrow.
5. Issue resolved: Great! Thank you!

Automate routine customer calls.

Voice for IT Support

Employee calls: My account is locked. The Copilot agent understands and helps.

1. Understand request: identify the issue using natural language.
2. Collect information: ask for employee ID and verification.
3. Trigger action: reset password or unlock account through a secure connector.
4. Resolve or escalate: confirm resolution or hand over to a human agent.

Turn repetitive IT calls into automated workflows. Faster help. Happier employees. More productive teams.

Great Voice Agent Scenarios

• Customer service: answer common questions by phone.
• Order status: look up shipments and deliveries.
• Appointments: book, reschedule, or confirm times.
• Billing: explain charges and payment status.
• IT help desk: unlock accounts and guide fixes.
• Contact center: triage calls and route to specialists.

Start with high-volume, repeatable conversations.

Voice Needs Voice Testing

Test how people actually speak - not just ideal scenarios.

• Speech paths: test common and varied phrasings.
• DTMF inputs: test key presses such as 1, 2, 3.
• Silence: test no input and pauses.
• Unrecognized speech: test unclear or out-of-scope input.
• Conversation branches: test different user journeys.
• Escalation: test handoff to a human agent.

Real conversations. Real confidence.

From Build to Production

1. Build agent: create and configure your agent.
2. Enable voice: turn on the voice channel in Copilot Studio.
3. Configure voice: set voice, language and conversation behavior.
4. Configure telephony: connect your telephony provider.
5. Test: try real calls and validate scenarios.
6. Publish: publish your voice agent.
7. Voice workstream: monitor, improve and scale.

Configure. Test. Then go live. Build today for real impact tomorrow.

How Everything Connects

1. Caller: customer or employee.
2. Voice and telephony: PSTN or Teams telephony.
3. Copilot Studio agent: voice-enabled agent.
4. Topics, Knowledge, Actions: conversation logic, your content, take action.
5. Business systems: CRM, ERP, ITSM, databases, APIs.

Voice connects conversation to business systems.

Design for Voice, Not Chat

Different channel. Different experience. People speak differently than they type.

Chat (text-heavy): To check your order status, please provide your order number, the email address associated with the order, and the date of purchase. You can also choose from the following options: 1. Track order, 2. Cancel order, 3. Return item, 4. Speak to an agent.

Voice (simple and natural): Whats the status of my order?

• Short: keep it concise.
• Clear: use simple language.
• Conversational: natural and human-like.
• Easy to answer: ask one thing at a time.

Avoid long prompts and complex menus. People do not want to listen to a wall of text. Voice UX needs its own design.

Production Checklist

• Voice scenario, owner, audience, and success metrics are documented.
• Basic Voice or Real-time Voice is selected for the right use case.
• Agent, topics, knowledge, and actions are tested for voice phrasings.
• Voice output, language, tone, greetings, and prompts are configured.
• DTMF keypad options are mapped and tested (1 Sales, 2 Support, 3 Billing).
• Silence timeout, retry prompts, and escalation paths are configured.
• Barge-in and turn-taking behavior is validated.
• Telephony provider, phone numbers, and inbound/outbound calling are connected.
• Business data access, actions, and permission checks are secured.
• Speech paths, branches, unrecognized input, and handoff are tested with real calls.
• Monitoring covers call volume, fallback, action health, escalation, and satisfaction.
• Release notes, support ownership, and improvement rhythm are assigned.

Key Takeaway

Move beyond chat. Build agents that can talk. A voice-enabled Copilot Studio agent understands speech, accepts keypad input, responds naturally, completes tasks, and escalates when needed.

When voice, telephony, topics, knowledge, actions, business systems, testing, and human handoff work together, phone calls become real business outcomes. Next: Day 57: Direct Line Channel Explained for Custom Copilot Apps.