🎙️ VoiceFlow AI: Telegram + Deepgram + OpenAI + Supabase Audio Assistant

🎙️ 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 Deepgram STT API Deepgram TTS API Telegram Bot API OpenAI API Supabase Documentation

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.