🎙️ AI Audio Assistant with Voice-to-Voice Response
Who is this for?
Businesses, customer service teams, content creators, and organizations who want to provide intelligent voice-based interactions through Telegram. Perfect for accessibility-focused services, multilingual support, or hands-free customer assistance.
What problem does this solve?
- Enables natural voice conversations with AI
- Breaks down language and accessibility barriers
- Provides instant voice responses to customer queries
- Reduces typing requirements for users
- Offers 24/7 voice-based customer support
- Maintains conversation context across voice interactions
What this workflow does:
- Receives voice messages via Telegram bot
- Transcribes audio using Deepgram's advanced speech-to-text
- Processes transcribed text through AI agent with knowledge base access
- Generates intelligent responses based on conversation context
- Converts AI response to natural-sounding speech using Deepgram TTS
- Sends audio response back to user via Telegram
- Maintains conversation memory for contextual interactions
🔧 Technical Architecture
Core Components:
-
Telegram Bot: Receives and sends voice messages
-
Deepgram STT: Transcribes voice to text with high accuracy
-
OpenAI GPT: Processes queries and generates responses
-
Supabase Knowledge Base: Stores and retrieves business information
-
Memory Management: Maintains conversation context
-
Deepgram TTS: Converts text responses to natural speech
Data Flow:
- Voice Message → Telegram API → File Download
- Audio File → Deepgram STT → Transcript
- Transcript → AI Agent → Response Generation
- Response → Deepgram TTS → Audio File
- Audio Response → Telegram → User
🛠️ Setup Instructions
Prerequisites
-
Telegram Bot Token
- Create bot via @BotFather
- Get bot token and configure webhook
-
Deepgram API Key
- Sign up at deepgram.com
- Get API key for STT and TTS services
- Note: Currently hardcoded in workflow
-
OpenAI API Key
- OpenAI account with API access
- Configure in OpenAI Chat Model node
-
Supabase Database
- Create Supabase project
- Set up knowledge_base table
- Configure API credentials
Step-by-Step Setup
-
Configure Telegram Bot
- Update telegramToken in "Prepare Voice Message Data" node
- Set correct bot token in Telegram nodes
- Test bot connectivity
-
Set Up Deepgram Integration
- Replace API key in "Transcribe with Deepgram" node
- Update TTS endpoint in "HTTP Request" node
- Test voice transcription accuracy
-
Configure Knowledge Base
-- Create knowledge_base table in Supabase
CREATE TABLE knowledge_base (
id UUID DEFAULT gen_random_uuid() PRIMARY KEY,
question TEXT NOT NULL,
answer TEXT NOT NULL,
category VARCHAR(100),
keywords TEXT[],
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
-
Customize AI Prompts
- Update system message in "Telegram AI Agent" node
- Adjust temperature and max tokens in OpenAI model
- Configure memory session keys
-
Test End-to-End Flow
- Send test voice message to bot
- Verify transcription accuracy
- Check AI response quality
- Validate audio output clarity
🎛️ Configuration Options
Voice Recognition Settings
-
Model: nova-2 (Deepgram's latest model)
-
Language: English (en) - can be changed
-
Smart Format: Enabled for better punctuation
AI Response Settings
-
Temperature: 0.3 (conservative responses)
-
Max Tokens: 100 (adjust based on needs)
-
Memory: Session-based conversation context
Text-to-Speech Settings
-
Model: aura-2-thalia-en (natural female voice)
-
Alternative voices: Available in Deepgram TTS API
-
Audio Format: Optimized for Telegram
🔒 Security Considerations
API Key Management
// Current implementation has hardcoded tokens
// Recommended: Use environment variables
const telegramToken = process.env.TELEGRAM_BOT_TOKEN;
const deepgramKey = process.env.DEEPGRAM_API_KEY;
Data Privacy
- Voice messages are processed by external APIs
- Consider data retention policies
- Implement user consent mechanisms
- Ensure GDPR compliance if applicable
📊 Monitoring & Analytics
Key Metrics to Track
- Voice message processing time
- Transcription accuracy rates
- AI response quality scores
- User engagement metrics
- Error rates and failure points
Recommended Logging
// Add to workflow for monitoring
console.log({
timestamp: new Date().toISOString(),
user_id: userData.user_id,
transcript_confidence: transcriptData.confidence,
response_length: aiResponse.length,
processing_time: processingTime
});
🚀 Customization Ideas
Enhanced Features
-
Multi-language Support
- Add language detection
- Support multiple TTS voices
- Translate responses
-
Voice Commands
- Implement wake words
- Add voice shortcuts
- Create voice menus
-
Advanced AI Features
- Sentiment analysis
- Intent classification
- Escalation triggers
-
Integration Expansions
- Connect to CRM systems
- Add calendar scheduling
- Integrate with help desk tools
Performance Optimizations
- Implement audio preprocessing
- Add response caching
- Optimize API call sequences
- Implement retry mechanisms
🐛 Troubleshooting
Common Issues
Voice Not Transcribing
- Check Deepgram API key validity
- Verify audio format compatibility
- Test with shorter voice messages
Poor Audio Quality
- Adjust TTS model settings
- Check network connectivity
- Verify Telegram audio limits
AI Responses Too Generic
- Improve knowledge base content
- Adjust system prompts
- Increase context window
Memory Not Working
- Check session key configuration
- Verify user ID extraction
- Test conversation continuity
💡 Best Practices
Voice Interface Design
- Keep responses concise and clear
- Use natural speech patterns
- Avoid technical jargon
- Provide clear next steps
Knowledge Base Management
- Regular content updates
- Clear categorization
- Keyword optimization
- Quality assurance testing
User Experience
- Fast response times (<5 seconds)
- Consistent voice personality
- Graceful error handling
- Clear capability communication
📈 Success Metrics
Technical KPIs
- Response time: <3 seconds average
- Transcription accuracy: >95%
- User satisfaction: >4.5/5
- Uptime: >99.5%
Business KPIs
- Customer query resolution rate
- Support ticket reduction
- User engagement increase
- Cost per interaction decrease
🔄 Maintenance Schedule
Daily
- Monitor error logs
- Check API rate limits
- Verify service uptime
Weekly
- Review conversation quality
- Update knowledge base
- Analyze usage patterns
Monthly
- Performance optimization
- Security audit
- Feature updates
- User feedback review
📚 Additional Resources
Documentation Links
Community Support
- n8n Community Forum
- Telegram Bot Developers Group
- Deepgram Developer Discord
- OpenAI Developer Community
Note: This template requires active API subscriptions for Deepgram and OpenAI services. Costs may apply based on usage volume.
