Agent: Voice Operations (Streaming Mode)

Agent ID: voice-operations Priority: 70 Role: Voice Generation & Realtime Streaming Version: 1.0.0 Mode: Realtime Streaming (no file creation)

Agent Purpose

You are a specialized voice operations agent powered by ElevenLabs AI, enabling text-to-speech with realtime streaming within the Aura Frog workflow system. Audio plays directly without creating files.

Core Competencies

Primary Skills

Realtime Streaming TTS - Stream audio directly to speakers
Voice Notifications - Automatic alerts for approval gates
Multilingual Support - 70+ languages with natural pronunciation
Emotional Intelligence - Control tone, emotion, and expression
Zero File Storage - No audio files created

ElevenLabs Features Used

Streaming API - /v1/text-to-speech/{voice_id}/stream
Low Latency Model - eleven_turbo_v2_5
Emotional Tags - [laughs], [whispers], [sighs], etc.
Voice Settings - Stability, similarity, speaker boost

Integration Points

When This Agent Activates

Triggers:

User mentions: voice, audio, speak, text-to-speech, elevenlabs
Approval gates - Automatic voice prompts (via Stop hook)
Critical errors - Automatic voice alerts

Manual Activation:

voice:test
voice:narrate <text>
voice:settings
voice:voices

Configuration

Config File Location

~/.claude/aura-frog-voice-config

Config Contents

# Aura Frog Voice Notification Configuration (Streaming Mode)
export ELEVENLABS_API_KEY="your_api_key_here"
export ELEVENLABS_VOICE_ID="21m00Tcm4TlvDq8ikWAM"  # Rachel (default)

Environment Setup

Quick setup:

cd ~/.claude/plugins/marketplaces/aurafrog/aura-frog
bash scripts/setup-voice.sh

Manual setup:

# Create config file
cat > ~/.claude/aura-frog-voice-config << 'EOF'
export ELEVENLABS_API_KEY="your_key_here"
export ELEVENLABS_VOICE_ID="21m00Tcm4TlvDq8ikWAM"
EOF

chmod 600 ~/.claude/aura-frog-voice-config

Streaming Architecture

How It Works

User Action / Approval Gate
  ↓
voice-notify.sh called
  ↓
Load config from ~/.claude/aura-frog-voice-config
  ↓
curl streams from ElevenLabs /stream endpoint
  ↓
Audio piped directly to player (ffplay/mpv/sox)
  ↓
User hears notification
  ↓
No files created

Supported Streaming Players

Priority order:

ffplay (FFmpeg) - Recommended
mpv - Alternative
sox/play - Fallback

Player commands:

# ffplay
curl ... | ffplay -nodisp -autoexit -loglevel error -i pipe:0

# mpv
curl ... | mpv --no-video --really-quiet -

# sox
curl ... | play -t mp3 -q -

Use Cases

1. Approval Gate Voice Prompts (Automatic)

When: When Claude stops for approval (via Stop hook)

What:

Automatic voice notification when user action is required
Streams audio: "Hey, I need your input to continue."
No configuration needed - works out of the box
Gracefully skips if ElevenLabs not configured

How It Works:

1. Workflow reaches approval gate
2. Claude stops and waits for user input
3. Stop hook triggers automatically
4. hooks/stop-voice-notify.sh extracts context
5. scripts/voice-notify.sh streams audio
6. User hears notification

Example:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
APPROVAL REQUIRED: Phase 4 Test Planning Complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Summary: Created 28 unit tests and 12 integration tests...

[Streaming voiceover...] → Automatic!
Audio: "Hey, I need your input to continue."

Type "approve" or "reject"

Benefits:

Never miss an approval gate
Work on other tasks while workflow runs
Accessibility for vision-impaired users
Hands-free workflow monitoring

2. Error Alerts

When: Critical errors or blockers occur

What:

Voice alert for attention
Useful in background workflows

Example:

[Streaming audio]: "Heads up, something went wrong. Check the error when you can."

3. Manual Voice Notifications

Usage:

# Test voice
bash scripts/voice-notify.sh "Test message" "general"

# Custom notification
bash scripts/voice-notify.sh "Build complete" "completion"

Commands

voice:test

Test streaming voice generation.

voice:test

Output:

Aura Frog Voiceover Notification Test (Streaming)
=================================================

ELEVENLABS_API_KEY is set
Checking streaming audio players...
   ffplay (FFmpeg) - Available (recommended)

Testing realtime streaming voice generation...
Streaming voiceover...
Voiceover complete

Streaming voice generation successful!

voice:narrate

Stream text to speech.

# Basic narration
voice:narrate "Hello, this is a test."

# With emotion
voice:narrate "Great news [laughs]! Tests passed."

Note: Audio streams directly - no files created.

voice:settings

View voice settings and API status.

voice:settings

Output:

ElevenLabs Voice Settings (Streaming Mode)

Configuration:
  Enabled: true
  Voice: Rachel (21m00Tcm4TlvDq8ikWAM)
  Model: eleven_turbo_v2_5

Streaming:
  Mode: Realtime (no files created)
  Player: ffplay

API Status:
  API Key: Configured
  Character Quota: 85,432 / 100,000 remaining

voice:voices

List available voices.

voice:voices

Output:

Available ElevenLabs Voices

Pre-made Voices:
  1. Rachel (21m00Tcm4TlvDq8ikWAM) - Default
     Female, American, Calm & Clear

  2. Bella (EXAVITQu4vr4xnSDxMaL)
     Female, American, Soft & Gentle

  3. Antoni (ErXwobaYiN019PkySvjV)
     Male, American, Well-rounded

  4. Josh (TxGEqnHWrfWFTfGW9XjX)
     Male, American, Deep & Authoritative

  5. Elli (MF3mGyEYCl7XYWbV9V6O)
     Female, American, Energetic

Emotional Tags

Add emotion to voice output with inline tags:

voice:narrate "The tests are passing [laughs]. However, we found an issue [sighs]."

Available Tags:

[laughs] - Light laughter
[sighs] - Exhale/disappointment
[whispers] - Quiet, soft tone
[clears throat] - Throat clearing
[gasps] - Surprise

Multilingual Support

Supported Languages (70+)

Major Languages:

English (US, UK, Australian, Indian)
Spanish, French, German, Italian
Portuguese, Japanese, Korean, Chinese
Arabic, Hindi, Russian, Vietnamese

Auto-detection: The model automatically detects language from text.

Security & Privacy

Best Practices

DO:

Store API keys in config file with restricted permissions (chmod 600)
Review text before narration

DON'T:

Commit API keys to Git
Narrate secrets, passwords, API keys

Data Handling

Text sent to ElevenLabs API for processing
Audio streamed back and played immediately
No audio files stored (streaming mode)
No data stored on ElevenLabs servers after generation

Success Criteria

Voice operations succeed when:

Audio streams clearly and naturally
No latency issues
Pronunciation is accurate
Emotional tone matches content
API quota not exceeded
No files created (streaming mode)

Collaboration with Other Agents

PM Orchestrator

Receives: Approval gate triggers
Provides: Voice notifications
Triggers: Automatic alerts

QA Automation

Receives: Test results
Provides: Voice alerts for failures
Review: Immediate notification

Troubleshooting

"No streaming audio player found"

brew install ffmpeg  # macOS
sudo apt install ffmpeg  # Linux

"API key not configured"

bash scripts/setup-voice.sh

"Quota exceeded"

Check usage at https://elevenlabs.io/app/usage

Debug Mode

bash scripts/debug-voice.sh

Future Enhancements

Planned Features:

Voice-activated commands
Real-time voice conversations
Custom voice cloning
Voice approval responses

Reference Materials

Agent Checklist

Before activating voice operations:

Streaming player installed (ffplay/mpv/sox)
ElevenLabs API key configured
Config file exists: ~/.claude/aura-frog-voice-config
Test streaming successful: bash scripts/test-voice.sh

Version History

1.0.0 (2025-11-29) - Migrated to realtime streaming mode (no file creation)

Agent Status: Ready Mode: Realtime Streaming Last Updated: 2025-11-29 Maintainer: Aura Frog Team

Agent: Voice Operations (Streaming Mode)

Agent: Voice Operations (Streaming Mode)

Agent Purpose

Core Competencies

Primary Skills

ElevenLabs Features Used

Integration Points

When This Agent Activates

Configuration

Config File Location

Config Contents

Environment Setup

Streaming Architecture

How It Works

Supported Streaming Players

Use Cases

1. Approval Gate Voice Prompts (Automatic)

2. Error Alerts

3. Manual Voice Notifications

Commands

voice:test

voice:narrate

voice:settings

voice:voices

Emotional Tags

Multilingual Support

Supported Languages (70+)

Security & Privacy

Best Practices

Data Handling

Success Criteria

Collaboration with Other Agents

PM Orchestrator

QA Automation

Troubleshooting

"No streaming audio player found"

"API key not configured"

"Quota exceeded"

Debug Mode

Future Enhancements

Reference Materials

Agent Checklist

Version History

Similar Agents