The workflow JSON
Copy or download the full n8n JSON below. Paste it into a new n8n workflow, add your credentials, activate. Full import guide →
{
"name": "Voice AI Agent - Real-time Processing",
"nodes": [
{
"parameters": {
"path": "voice-stream",
"authentication": "jwtAuth",
"options": {
"binaryPropertyName": "audioStream",
"rawBody": true
}
},
"id": "a1b2c3d4-e5f6-7890-real-time-voice-input",
"name": "WebSocket Voice Input",
"type": "n8n-nodes-base.webhook",
"typeVersion": 2,
"position": [
240,
300
]
},
{
"parameters": {
"jsCode": "// Enhanced audio validation and preprocessing\nconst audioData = $binary.audioStream;\nconst sessionId = $json.sessionId || `session_${Date.now()}`;\nconst userId = $json.userId || 'anonymous';\n\n// Validate audio stream\nif (!audioData || audioData.fileSize === 0) {\n throw new Error('Invalid audio stream');\n}\n\n// Audio format validation\nconst supportedFormats = ['audio/wav', 'audio/webm', 'audio/ogg'];\nif (!supportedFormats.includes(audioData.mimeType)) {\n throw new Error(`Unsupported audio format: ${audioData.mimeType}`);\n}\n\n// Generate unique interaction ID with timestamp\nconst interactionId = `${userId}_${sessionId}_${Date.now()}`;\nconst timestamp = new Date().toISOString();\n\n// Audio stream metadata\nconst audioMetadata = {\n interactionId,\n sessionId,\n userId,\n timestamp,\n audioFormat: audioData.mimeType,\n audioSize: audioData.fileSize,\n sampleRate: $json.sampleRate || 16000,\n channels: $json.channels || 1,\n duration: audioData.fileSize / (16000 * 2), // Approximate duration for 16kHz 16-bit\n clientInfo: {\n userAgent: $json.userAgent,\n language: $json.language || 'en-US',\n timezone: $json.timezone || 'UTC'\n }\n};\n\nreturn {\n json: audioMetadata,\n binary: {\n audioStream: audioData\n }\n};"
},
"id": "b2c3d4e5-f6g7-8901-audio-preprocessing",
"name": "Enhanced Audio Preprocessing",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
460,
300
]
},
{
"parameters": {
"jsCode": "// Voice Activity Detection (VAD)\nconst audioData = $binary.audioStream;\nconst audioBuffer = Buffer.from(audioData.data, 'base64');\n\n// Simple energy-based VAD (in production, use WebRTC VAD or Silero VAD)\nfunction detectVoiceActivity(buffer) {\n const samples = new Int16Array(buffer.buffer);\n let energy = 0;\n \n for (let i = 0; i < samples.length; i++) {\n energy += samples[i] * samples[i];\n }\n \n const rms = Math.sqrt(energy / samples.length);\n const threshold = 500; // Adjust based on your needs\n \n return {\n isSpeech: rms > threshold,\n energy: rms,\n confidence: Math.min(rms / threshold, 1.0)\n };\n}\n\nconst vadResult = detectVoiceActivity(audioBuffer);\n\nreturn {\n json: {\n ...vadResult,\n timestamp: new Date().toISOString(),\n processAudio: vadResult.isSpeech\n },\n binary: {\n audioStream: $binary.audioStream\n }\n};"
},
"id": "c3d4e5f6-g7h8-9012-voice-activity-detection",
"name": "Voice Activity Detection",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
680,
300
]
},
{
"parameters": {
"conditions": {
"options": {
"combineOperation": "all"
},
"conditions": [
{
"id": "speech_detected",
"leftValue": "={{ $json.processAudio }}",
"rightValue": true,
"operation": "equal"
},
{
"id": "confidence_check",
"leftValue": "={{ $json.confidence }}",
"rightValue": 0.3,
"operation": "largerEqual"
}
]
}
},
"id": "d4e5f6g7-h8i9-0123-speech-validation",
"name": "Validate Speech Detection",
"type": "n8n-nodes-base.if",
"typeVersion": 2,
"position": [
900,
300
]
},
{
"parameters": {
"url": "http://whisper-server:8080/v1/audio/transcriptions",
"authentication": "genericCredentialType",
"genericAuthType": "httpHeaderAuth",
"sendBinaryData": true,
"binaryPropertyName": "audioStream",
"sendBody": true,
"bodyContentType": "multipart-form-data",
"bodyParameters": {
"parameters": [
{
"name": "model",
"value": "whisper-large-v3"
},
{
"name": "language",
"value": "={{ $('Enhanced Audio Preprocessing').item.json.clientInfo.language.split('-')[0] }}"
},
{
"name": "response_format",
"value": "verbose_json"
},
{
"name": "temperature",
"value": "0.0"
},
{
"name": "timestamp_granularities[]",
"value": "word"
}
]
},
"options": {
"timeout": 30000,
"retry": {
"enabled": true,
"maxTries": 3
}
}
},
"id": "e5f6g7h8-i9j0-1234-enhanced-stt",
"name": "Enhanced Speech-to-Text",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.2,
"position": [
1120,
240
],
"continueOnFail": true,
"retryOnFail": true,
"maxTries": 3
},
{
"parameters": {
"url": "http://embedding-service:8000/search",
"authentication": "none",
"sendBody": true,
"bodyContentType": "json",
"jsonBody": "={\n \"query\": \"{{ $json.text }}\",\n \"limit\": 5\n}",
"options": {
"timeout": 10000
}
},
"id": "f6g7h8i9-j0k1-2345-rag-search",
"name": "RAG Knowledge Search",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.2,
"position": [
1340,
240
],
"continueOnFail": true
},
{
"parameters": {
"jsCode": "// Enhanced context building with RAG results\nconst transcription = $('Enhanced Speech-to-Text').item.json;\nconst ragResults = $('RAG Knowledge Search').item.json || [];\nconst audioMetadata = $('Enhanced Audio Preprocessing').item.json;\n\n// Build conversation context\nconst userQuery = transcription.text;\nconst confidence = transcription.confidence || 0.0;\nconst language = transcription.language || 'en';\n\n// Extract relevant knowledge from RAG\nconst relevantKnowledge = ragResults\n .filter(result => result.score > 0.7)\n .map(result => result.text)\n .join('\\n\\n');\n\n// Build enhanced prompt with context\nconst systemPrompt = `You are a helpful AI assistant specialized in voice interactions.\n\nUser Information:\n- Language: ${language}\n- Session: ${audioMetadata.sessionId}\n- Interaction: ${audioMetadata.interactionId}\n\nRelevant Knowledge:\n${relevantKnowledge || 'No specific knowledge found for this query.'}\n\nInstructions:\n- Keep responses conversational and natural for voice output\n- If transcription confidence is low (< 0.8), ask for clarification\n- Use the provided knowledge when relevant\n- Respond in the same language as the user\n- Keep responses under 200 words for voice synthesis`;\n\nconst userPrompt = `User said: \"${userQuery}\"\nTranscription confidence: ${confidence}`;\n\nreturn {\n json: {\n systemPrompt,\n userPrompt,\n userQuery,\n confidence,\n language,\n hasKnowledge: relevantKnowledge.length > 0,\n knowledgeSnippets: ragResults.length,\n processingTimestamp: new Date().toISOString()\n }\n};"
},
"id": "g7h8i9j0-k1l2-3456-context-building",
"name": "Build Enhanced Context",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
1560,
240
]
},
{
"parameters": {
"url": "http://ollama-server:11434/v1/chat/completions",
"authentication": "none",
"sendBody": true,
"bodyContentType": "json",
"jsonBody": "={\n \"model\": \"{{ $vars.LLM_MODEL || 'llama3.2:8b' }}\",\n \"messages\": [\n {\n \"role\": \"system\",\n \"content\": \"{{ $json.systemPrompt }}\"\n },\n {\n \"role\": \"user\",\n \"content\": \"{{ $json.userPrompt }}\"\n }\n ],\n \"temperature\": 0.7,\n \"max_tokens\": 300,\n \"top_p\": 0.9,\n \"presence_penalty\": 0.1,\n \"frequency_penalty\": 0.1\n}",
"options": {
"timeout": 30000
}
},
"id": "h8i9j0k1-l2m3-4567-llm-processing",
"name": "Enhanced LLM Processing",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.2,
"position": [
1780,
240
],
"continueOnFail": true,
"retryOnFail": true,
"maxTries": 2
},
{
"parameters": {
"jsCode": "// Post-process LLM response for voice optimization\nconst llmResponse = $json.choices?.[0]?.message?.content || $json.response || $json.text || '';\nconst metadata = $('Build Enhanced Context').item.json;\n\n// Clean and optimize response for TTS\nfunction optimizeForVoice(text) {\n return text\n .replace(/\\*\\*(.*?)\\*\\*/g, '$1') // Remove markdown bold\n .replace(/\\*(.*?)\\*/g, '$1') // Remove markdown italic\n .replace(/```[\\s\\S]*?```/g, '') // Remove code blocks\n .replace(/\\[([^\\]]+)\\]\\([^)]+\\)/g, '$1') // Convert links to text\n .replace(/\\s+/g, ' ') // Normalize whitespace\n .trim();\n}\n\nconst optimizedResponse = optimizeForVoice(llmResponse);\n\n// Add SSML tags for better speech synthesis\nconst ssmlResponse = `<speak>\n <prosody rate=\"medium\" pitch=\"medium\">\n ${optimizedResponse}\n </prosody>\n</speak>`;\n\nreturn {\n json: {\n originalResponse: llmResponse,\n optimizedResponse,\n ssmlResponse,\n responseLength: optimizedResponse.length,\n estimatedSpeechDuration: Math.ceil(optimizedResponse.length / 15), // ~15 chars per second\n processingComplete: true,\n metadata: {\n interactionId: metadata.interactionId || 'unknown',\n language: metadata.language,\n hasKnowledge: metadata.hasKnowledge,\n confidence: metadata.confidence\n }\n }\n};"
},
"id": "i9j0k1l2-m3n4-5678-response-optimization",
"name": "Optimize Response for Voice",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
2000,
240
]
},
{
"parameters": {
"url": "http://kokoro-tts:8880/v1/audio/speech",
"authentication": "none",
"sendBody": true,
"bodyContentType": "json",
"jsonBody": "={\n \"model\": \"kokoro-v2\",\n \"input\": \"{{ $json.ssmlResponse }}\",\n \"voice\": \"{{ $vars.TTS_VOICE || 'af_bella' }}\",\n \"response_format\": \"mp3\",\n \"speed\": 1.0,\n \"pitch\": 0.0,\n \"voice_settings\": {\n \"stability\": 0.75,\n \"similarity_boost\": 0.75,\n \"style\": 0.5\n }\n}",
"options": {
"timeout": 30000,
"response": {
"responseFormat": "file",
"outputPropertyName": "audioResponse"
}
}
},
"id": "j0k1l2m3-n4o5-6789-enhanced-tts",
"name": "Enhanced Text-to-Speech",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.2,
"position": [
2220,
240
],
"continueOnFail": true,
"retryOnFail": true,
"maxTries": 3
},
{
"parameters": {
"assignments": {
"assignments": [
{
"id": "interaction_id",
"name": "interactionId",
"value": "={{ $('Enhanced Audio Preprocessing').item.json.interactionId }}",
"type": "string"
},
{
"id": "session_id",
"name": "sessionId",
"value": "={{ $('Enhanced Audio Preprocessing').item.json.sessionId }}",
"type": "string"
},
{
"id": "user_id",
"name": "userId",
"value": "={{ $('Enhanced Audio Preprocessing').item.json.userId }}",
"type": "string"
},
{
"id": "user_input",
"name": "userInput",
"value": "={{ $('Enhanced Speech-to-Text').item.json.text }}",
"type": "string"
},
{
"id": "ai_response",
"name": "aiResponse",
"value": "={{ $('Optimize Response for Voice').item.json.optimizedResponse }}",
"type": "string"
},
{
"id": "confidence",
"name": "confidence",
"value": "={{ $('Enhanced Speech-to-Text').item.json.confidence }}",
"type": "number"
},
{
"id": "language",
"name": "language",
"value": "={{ $('Enhanced Speech-to-Text').item.json.language }}",
"type": "string"
},
{
"id": "processing_time",
"name": "processingTime",
"value": "={{ new Date().getTime() - new Date($('Enhanced Audio Preprocessing').item.json.timestamp).getTime() }}",
"type": "number"
},
{
"id": "rag_used",
"name": "ragUsed",
"value": "={{ $('Optimize Response for Voice').item.json.metadata.hasKnowledge }}",
"type": "boolean"
},
{
"id": "audio_format",
"name": "audioFormat",
"value": "mp3",
"type": "string"
},
{
"id": "status",
"name": "status",
"value": "completed",
"type": "string"
},
{
"id": "timestamp",
"name": "timestamp",
"value": "={{ new Date().toISOString() }}",
"type": "string"
}
]
},
"options": {}
},
"id": "k1l2m3n4-o5p6-7890-format-response",
"name": "Format Enhanced Response",
"type": "n8n-nodes-base.set",
"typeVersion": 3.4,
"position": [
2440,
240
]
},
{
"parameters": {
"respondWith": "json",
"responseBody": "={\n \"success\": true,\n \"interactionId\": \"{{ $json.interactionId }}\",\n \"response\": {\n \"text\": \"{{ $json.aiResponse }}\",\n \"audio\": {\n \"format\": \"{{ $json.audioFormat }}\",\n \"data\": \"{{ $binary.audioResponse ? Buffer.from($binary.audioResponse.data).toString('base64') : null }}\"\n }\n },\n \"metadata\": {\n \"confidence\": {{ $json.confidence }},\n \"language\": \"{{ $json.language }}\",\n \"processingTime\": {{ $json.processingTime }},\n \"ragUsed\": {{ $json.ragUsed }},\n \"timestamp\": \"{{ $json.timestamp }}\"\n }\n}",
"options": {
"responseHeaders": {
"entries": [
{
"name": "Content-Type",
"value": "application/json"
},
{
"name": "X-Voice-AI-Version",
"value": "2.0"
}
]
}
}
},
"id": "l2m3n4o5-p6q7-8901-webhook-response",
"name": "Send Enhanced Response",
"type": "n8n-nodes-base.respondToWebhook",
"typeVersion": 1.1,
"position": [
2660,
240
]
},
{
"parameters": {
"workflowId": "{{ $vars.ENHANCED_LOGGING_WORKFLOW_ID }}",
"rawInputs": {
"workflowInputs": [
{
"name": "interactionData",
"value": "={{ JSON.stringify($json) }}"
},
{
"name": "audioMetadata",
"value": "={{ JSON.stringify($('Enhanced Audio Preprocessing').item.json) }}"
},
{
"name": "vadResults",
"value": "={{ JSON.stringify($('Voice Activity Detection').item.json) }}"
},
{
"name": "transcriptionData",
"value": "={{ JSON.stringify($('Enhanced Speech-to-Text').item.json) }}"
}
]
},
"options": {
"waitForCompletion": false
}
},
"id": "m3n4o5p6-q7r8-9012-enhanced-logging",
"name": "Enhanced Interaction Logging",
"type": "n8n-nodes-base.executeWorkflow",
"typeVersion": 1.2,
"position": [
2440,
400
],
"continueOnFail": true
},
{
"parameters": {
"respondWith": "json",
"responseBody": "={\n \"error\": true,\n \"message\": \"No speech detected or confidence too low\",\n \"details\": {\n \"speechDetected\": {{ $('Voice Activity Detection').item.json.isSpeech }},\n \"confidence\": {{ $('Voice Activity Detection').item.json.confidence }},\n \"energy\": {{ $('Voice Activity Detection').item.json.energy }}\n },\n \"suggestions\": [\n \"Speak closer to the microphone\",\n \"Reduce background noise\",\n \"Speak more clearly\"\n ]\n}",
"options": {
"responseCode": 200,
"responseHeaders": {
"entries": [
{
"name": "Content-Type",
"value": "application/json"
}
]
}
}
},
"id": "n4o5p6q7-r8s9-0123-no-speech-response",
"name": "No Speech Detected Response",
"type": "n8n-nodes-base.respondToWebhook",
"typeVersion": 1.1,
"position": [
900,
500
]
}
],
"connections": {
"WebSocket Voice Input": {
"main": [
[
{
"node": "Enhanced Audio Preprocessing",
"type": "main",
"index": 0
}
]
]
},
"Enhanced Audio Preprocessing": {
"main": [
[
{
"node": "Voice Activity Detection",
"type": "main",
"index": 0
}
]
]
},
"Voice Activity Detection": {
"main": [
[
{
"node": "Validate Speech Detection",
"type": "main",
"index": 0
}
]
]
},
"Validate Speech Detection": {
"main": [
[
{
"node": "Enhanced Speech-to-Text",
"type": "main",
"index": 0
}
],
[
{
"node": "No Speech Detected Response",
"type": "main",
"index": 0
}
]
]
},
"Enhanced Speech-to-Text": {
"main": [
[
{
"node": "RAG Knowledge Search",
"type": "main",
"index": 0
}
]
]
},
"RAG Knowledge Search": {
"main": [
[
{
"node": "Build Enhanced Context",
"type": "main",
"index": 0
}
]
]
},
"Build Enhanced Context": {
"main": [
[
{
"node": "Enhanced LLM Processing",
"type": "main",
"index": 0
}
]
]
},
"Enhanced LLM Processing": {
"main": [
[
{
"node": "Optimize Response for Voice",
"type": "main",
"index": 0
}
]
]
},
"Optimize Response for Voice": {
"main": [
[
{
"node": "Enhanced Text-to-Speech",
"type": "main",
"index": 0
}
]
]
},
"Enhanced Text-to-Speech": {
"main": [
[
{
"node": "Format Enhanced Response",
"type": "main",
"index": 0
}
]
]
},
"Format Enhanced Response": {
"main": [
[
{
"node": "Send Enhanced Response",
"type": "main",
"index": 0
},
{
"node": "Enhanced Interaction Logging",
"type": "main",
"index": 0
}
]
]
}
},
"settings": {
"executionOrder": "v1",
"saveManualExecutions": true,
"callerPolicy": "workflowsFromSameOwner",
"errorWorkflow": "{{ $vars.ERROR_HANDLING_WORKFLOW_ID }}"
},
"staticData": {},
"tags": [
"voice-ai",
"enhanced",
"real-time"
],
"triggerCount": 1,
"updatedAt": "2025-06-27T00:00:00.000Z",
"versionId": "2.0"
}
Pro
For the full experience including quality scoring and batch install features for each workflow upgrade to Pro
About this workflow
Voice AI Agent - Real-time Processing. Uses httpRequest. Webhook trigger; 14 nodes.
Source: https://github.com/161sam/n8n-voice-ai/blob/main/workflows/Voice-AI-Agent-Real-time-Processing.json — original creator credit. Request a take-down →
Related workflows
Workflows that share integrations, category, or trigger type with this one. All free to copy and import.
Jigsaw API key for image processing, I use this as a gatekeeper/second pair of eyes. LINK to their website https://jigsawstack.com/ SECOND A postgress DATABASE (I use Supabase) LlamaCloud for the pars
HTTP Request, Postgres, Stop And Error +2
Whatsapp Multi Agent System optimized copy 2.0. Uses airtable, httpRequest, errorTrigger. Webhook trigger; 44 nodes.
Airtable, HTTP Request, Error Trigger
Invoice Agent. Uses httpRequest, emailSend. Webhook trigger; 29 nodes.
HTTP Request, Email Send
Reputation Engine — SEO QA Agent. Uses httpRequest. Webhook trigger; 28 nodes.
HTTP Request
This workflow handles incoming voice calls or audio messages, transcribes them using Whisper (OpenAI) or ElevenLabs, extracts booking intent and preferred time slots using AI, checks availability on C
HTTP Request