| name | agent-voice-configuration |
| description | Build voice-enabled agents in watsonx Orchestrate using the ADK. This guide covers initial setup, key concepts for building your first voice agent. |
Getting Started with Voice Agents
Overview
Quick start guide for building voice-enabled agents in watsonx Orchestrate using the ADK. This guide covers initial setup, key concepts, and your first voice agent.
Prerequisites
- Python 3.9 or higher
- watsonx Orchestrate instance access
- Basic understanding of agent concepts (see agent-builder skill)
ADK Installation
Standard Installation
pip install ibm-watsonx-orchestrate
Installs base ADK without voice capabilities.
Voice-Enabled Installation (Required for Voice Agents)
pip install ibm-watsonx-orchestrate --with-voice
CRITICAL: The --with-voice flag is REQUIRED for voice agent development.
Verify Installation
wxo --version
Core ADK Principle
ALWAYS use watsonx Orchestrate ADK MCP tools to build agents directly in the platform.
ALWAYS search the ADK documentation BEFORE creating or modifying any resources.
The ADK documentation is the source of truth for current specifications and best practices.
Before Creating Any Asset
Search ADK documentation for:
- Best practices for the resource type
- Naming conventions and standards
- Required and optional fields
- Common patterns and examples
This ensures consistency and follows platform standards.
Key ADK Tools for Voice Agents
Discovery Tools
list_agents - See existing agents
list_voice_configs - Check voice configurations
list_channels - View channel integrations
list_phone_configs - Check phone configurations
list_tools - Understand available tools
Creation Tools
create_or_update_agent - Build agents in Orchestrate
import_voice_config - Set up voice configurations
create_or_update_channel - Configure communication channels
import_phone_config - Set up phone integrations
attach_agent_to_phone_config - Link agent to phone channel
Validation Tools
export_agent - Review agent specifications
generate_webchat_embed - Get webchat embed code
Voice Agent Components
1. Agent Definition
The core agent with instructions, tools, and collaborators. Voice agents need:
- Concise, conversational instructions
- Voice-optimized response guidelines
- Reference to voice configuration
2. Voice Configuration
Controls Speech-to-Text (STT) and Text-to-Speech (TTS):
- STT provider and model selection
- TTS provider and voice selection
- Audio quality parameters
- Language settings
3. Channel Integration
Connects agent to communication platforms:
- Phone: Genesys Audio Connector
- WhatsApp: Twilio integration
- SMS: Twilio integration
- Slack: BYO (Bring Your Own) integration
- Webchat: Built-in, automatically available
Your First Voice Agent
Step 1: Explore Environment
wxo agent list
wxo voice-config list
wxo channel list
Step 2: Search ADK Documentation
Before creating anything, search for:
- "voice agent architecture best practices"
- "voice configuration YAML specification"
- "agent instructions best practices"
Step 3: Create Voice Configuration
Search ADK docs, then create voice config YAML and import:
wxo voice-config import voice_config.yaml
Step 4: Create Agent
Search ADK docs, then use ADK MCP tool to create agent directly in Orchestrate with voice-optimized instructions.
Step 5: Configure Channel
Search ADK docs for channel type, then set up integration:
wxo phone-config import phone_config.yaml
wxo phone-config attach --config-name my-phone --agent-name my-agent
wxo channel import channel_config.yaml
Step 6: Test
Test voice interaction through configured channel and verify:
- STT accuracy
- TTS naturalness
- Response appropriateness
- Audio quality
File Creation Guidelines
When to Create Files
plan.md for user review before implementation
- YAML files temporarily before importing to Orchestrate
README.md only if user specifically requests documentation
- Sample credential templates
When NOT to Create Files
- Building agents - use
create_or_update_agent instead
- Creating voice configs - use
import_voice_config instead
- Setting up channels - use
create_or_update_channel instead
- Creating tools - use
create_tool or import_tool instead
Prefer using ADK MCP tools to build directly in watsonx Orchestrate rather than creating local files.
Greeting Template
When user greets you with "hi", "hello", "hey":
Hello! I'm Bob, your watsonx Orchestrate voice agent specialist. I help you build
voice-enabled agents directly in watsonx Orchestrate using the ADK.
**What I Can Help You With:**
**Building Voice Agents in watsonx Orchestrate:**
- Create new voice-enabled agents from scratch using ADK tools
- Add voice capabilities to existing agents
- Configure Speech-to-Text (STT) and Text-to-Speech (TTS)
- Set up voice channels (phone, WhatsApp, SMS, Slack)
**My Approach:**
- Use ADK MCP tools to build directly in watsonx Orchestrate
- Search ADK documentation for latest specifications
- Create comprehensive plans for your review
- Guide you through each step with clear explanations
What would you like to build today?
Next Steps
Once you understand the basics:
- Review
workflow-patterns.md for complete workflows
- Study
voice-configuration.md for STT/TTS setup
- Explore
channel-integration.md for channel-specific setup
- Learn
voice-optimization.md for instruction optimization
- Use
deployment.md for production deployment
Best Practices
- Always search ADK documentation before creating resources
- Create minimal files - only plan.md is required
- Use ADK MCP tools to build directly in Orchestrate
- Test voice quality early and often
- Keep responses concise (15-30 seconds when spoken)
- Use natural, conversational language
- Confirm critical information in voice interactions