Convert YouTube Videos to Audio Summaries with n8n

Send a link via Telegram → Decodo extracts transcript → OpenAI generates summary script and audio → Telegram returns MP3 and structured text

Who

Users who want audio summaries of YouTube videos via a Telegram bot

You get

Receive an MP3 audio summary plus a formatted message with genre tag, title, and text summary

You need

n8n · telegram · decodo

Steps

1. Configure workflow and credentials

   After importing the template into n8n, configure credentials: Telegram Bot Token, Decodo API Key, and OpenAI API Key. Double-click the Config node to set output_language (e.g., English, Spanish, German). The AI summary and speech will adapt to this language.

2. Trigger by sending a YouTube link via Telegram

   Send a YouTube video link to the connected Telegram bot (supports youtube.com/watch?v=... or youtu.be/... formats).

3. Validate URL with regex

   The workflow uses an If node with a regex to check whether the message contains a valid YouTube URL. If invalid, the bot replies with an error. If valid, it proceeds and sends a "Found it! 🎬 fetching transcript..." waiting message.

4. Fetch transcript with Decodo

   The Decodo node scrapes the full transcript JSON from the video page. Configure your Decodo API credential in the node and ensure the operation is set to youtube_transcript.

5. Check for transcript extraction errors

   An If node checks whether the Decodo result contains an error. If the video has no captions or extraction fails, the bot sends an error notification and the workflow stops.

   Execution prompt · Source text。Send Transcript Failed Message node text field

   Copy prompt

   =⚠️ **Transcript Failed**
   I couldn't extract text from this video. It might not have captions available.
   **Reason:**
   {{ $json.results[0].content.error.message }}

6. Format transcript with Code node

   The Code node flattens the deep JSON structure from Decodo into a clean text string for the AI to read.

   Execution prompt · Source text。Code: Format Transcript node JS Code field

   Copy prompt

   const content = $input.first().json.results?.[0]?.content;
   
   // Safety check: Ensure content is actually an array (valid transcript)
   // If it's an object (error) or null, return empty text.
   if (!Array.isArray(content)) {
       return [{ json: { transcript: "" } }];
   }
   
   const fullText = content.map(item => {
       return item.transcriptSegmentRenderer?.snippet?.runs?.[0]?.text || "";
   }).join(' ');
   
   return [{ json: { transcript: fullText } }];

7. AI Agent generates structured summary script

   The AI Summarizer node (Agent type) connects to an OpenAI Chat Model (gpt-4o-mini) and a Structured Output Parser. The user message passes the formatted transcript; the system prompt instructs the AI to output JSON in the configured language, containing genre, title, summary, and script (for TTS, under 300 words, must start with "This is a summary of the video...").

   Execution prompt · Source text。AI Summarizer node text / user message field

   Copy prompt

   =**Input Transcript:**
   {{ $json.transcript }}

   Execution prompt · Source text。AI Summarizer node options.systemMessage field

   Copy prompt

   =You are an expert Audio Summarizer. Your goal is to convert raw video transcripts into clear, concise, and easy-to-listen-to summaries.
   
   **CRITICAL INSTRUCTION:**
   You MUST write the entire summary, title, and script in **{{ $('Config').item.json.output_language }}**.
   
   **Phase 1: Analyze the Content**
   Determine the core subject matter (e.g., Cooking, Tech, History, News) and adopt a tone that fits:
   * **Educational:** Clear, instructional, and logical.
   * **News/Business:** Professional, objective, and direct.
   * **Story/Entertainment:** Engaging but descriptive.
   
   **Phase 2: Critical Content Extraction**
   **Do not miss the main point.** Identify the central argument, key facts, or the most valuable takeaway. The summary must deliver the value of the video without the fluff.
   
   **Phase 3: Script Guidelines**
   1.  **The Opening:** You MUST start the script with the phrase: "This is a summary of the video..." followed by the main topic or title.
   2.  **No Visual Cues:** The listener cannot see the video. Do NOT use phrases like "as you can see" or "in this chart." Rephrase them verbally.
   3.  **Direct & Clean:** Remove filler words, speaker labels, and conversational tangents. Just deliver the information.
   4.  **Strict Length:** The script must be **less than 300 words**.
   
   **Output Schema (Strict JSON):**
   You must output a single valid JSON object. Do not wrap it in markdown code blocks. Use this exact structure:
   {
     "genre": "The detected genre (e.g., 'True Crime', 'Cooking', 'Tech News')",
     "title": "A descriptive title for the file",
     "summary": "A short, punchy 2-sentence text caption for Telegram.",
     "script": "The full text for the Text-to-Speech engine to read. Start with 'This is a summary of the video...'."
   }

8. OpenAI TTS generates audio

   The OpenAI TTS node receives the script field from the AI output and calls the OpenAI audio endpoint to generate an MP3 file. Source notes mention using tts-1 for fast generation.

9. Telegram delivers audio and text summary

   The workflow sends two messages in sequence: first the MP3 audio file (Send Audio Summary), then a formatted HTML text summary (Send Text Summary) containing the title, genre tag, and summary.

   Execution prompt · Source text。Send Text Summary node text field (HTML parse_mode)

   Copy prompt

   =<b>{{ $('AI Summarizer').item.json.output.title }}</b>
   
   #{{ $('AI Summarizer').item.json.output.genre.replace(/ /g, '') }}
   
   <blockquote>{{ $('AI Summarizer').item.json.output.summary }}</blockquote>

You are an n8n workflow setup assistant. The user wants to recreate the "YouTube Video to Audio Summary" template. First confirm they have ready: an n8n instance, Telegram Bot Token, Decodo API Key, and OpenAI API Key. Then guide them to import the template, configure credentials, set output_language, and finally test by sending a YouTube link.