1// HTTP Request node: OpenAI TTS API
2// Method: POST
3// URL: https://api.openai.com/v1/audio/speech
4
5const ttsRequest = {
6  model: "tts-1",      // or "tts-1-hd" for higher quality
7  input: $json.text,
8  voice: "alloy",      // Options: alloy, echo, fable, onyx, nova, shimmer
9  response_format: "mp3", // mp3, opus, aac, flac
10  speed: 1.0           // 0.25 to 4.0
11};
12
13// Headers:
14// Authorization: Bearer $OPENAI_API_KEY
15// Content-Type: application/json
16
17// Response: Binary audio data (mp3)

1// Workflow: Convert text to speech and send via Telegram
2
3// Step 1: Prepare text (clean, add pauses)
4function prepareForTTS(text) {
5  return text
6    .replace(/\n\n/g, '... ')    // Paragraphs to pauses
7    .replace(/\n/g, '. ')         // Line breaks to periods
8    .replace(/[*_#]/g, '')        // Remove markdown formatting
9    .substring(0, 4096);          // TTS limit per request
10}
11
12// Step 2: Call TTS API (HTTP Request node)
13// Step 3: Send audio via Telegram (Telegram node: Send Audio)

1// HTTP Request node: ElevenLabs TTS
2// More natural voices, voice cloning supported
3
4const elevenLabsRequest = {
5  url: `https://api.elevenlabs.io/v1/text-to-speech/${voiceId}`,
6  method: "POST",
7  headers: {
8    "xi-api-key": process.env.ELEVENLABS_API_KEY,
9    "Content-Type": "application/json"
10  },
11  body: {
12    text: $json.text,
13    model_id: "eleven_multilingual_v2",
14    voice_settings: {
15      stability: 0.5,
16      similarity_boost: 0.75,
17      style: 0.5,
18      use_speaker_boost: true
19    }
20  }
21};
22
23// Response: Binary audio stream

1// Code node: Optimize text for speech
2function optimizeForSpeech(text) {
3  let speech = text;
4  
5  // Convert abbreviations to spoken form
6  speech = speech.replace(/API/g, 'A P I');
7  speech = speech.replace(/URL/g, 'U R L');
8  speech = speech.replace(/SQL/g, 'sequel');
9  
10  // Add natural pauses
11  speech = speech.replace(/\./g, '.\n');
12  speech = speech.replace(/:/g, '...');
13  
14  // Remove elements that don't sound natural
15  speech = speech.replace(/\[.*?\]/g, '');  // Remove citations
16  speech = speech.replace(/\|/g, ', ');     // Table pipes to commas
17  
18  return speech;
19}
20
21return { json: { speechText: optimizeForSpeech($json.aiResponse) } };

1// OpenAI TTS limit: 4096 characters per request
2// For longer texts: split into chunks
3
4function splitForTTS(text, maxChars = 4000) {
5  const sentences = text.match(/[^.!?]+[.!?]+/g) || [text];
6  const chunks = [];
7  let current = '';
8  
9  for (const sentence of sentences) {
10    if ((current + sentence).length > maxChars) {
11      chunks.push(current.trim());
12      current = sentence;
13    } else {
14      current += ' ' + sentence;
15    }
16  }
17  if (current) chunks.push(current.trim());
18  
19  return chunks;
20}
21
22// Process each chunk → Concatenate audio files
23const chunks = splitForTTS($json.longText);
24return chunks.map((c, i) => ({ json: { text: c, index: i } }));

1// Weekly: Convert newsletter to audio format
2// Schedule (weekly) → Fetch newsletter → Clean text → TTS → Upload to S3 → Send link
3
4const newsletterPrompt = `
5Convert this newsletter into a natural script for audio reading.
6
7Newsletter content:
8${$json.newsletterContent}
9
10Script rules:
11- Add intro: "Welcome to this week's newsletter"
12- Transition between sections: "Moving on to..."
13- Add outro: "That's all for this week"
14- Make it conversational, not robotic
15- Remove URLs and links (say "check the link in the show notes")
16`;