Generate Podcast from PDF using AI

git clone https://github.com/junwatu/griddb-podcast-creator.git
cd griddb-podcast-creator
cd apps
npm install

MISTRAL_API_KEY=
OPENAI_API_KEY=
GRIDDB_WEBAPI_URL=
GRIDDB_PASSWORD=
GRIDDB_USERNAME=

npm run dev

GRIDDB_WEBAPI_URL=
GRIDDB_USERNAME=
GRIDDB_PASSWORD=

export async function POST(request: NextRequest) {
  try {
    const formData = await request.formData();
    const file = formData.get('file') as File;

    if (!file) {
      return NextResponse.json(
        { error: 'No file uploaded' },
        { status: 400 }
      );
    }

    if (file.type !== 'application/pdf') {
      return NextResponse.json(
        { error: 'Invalid file type. Please upload a PDF file' },
        { status: 400 }
      );
    }

    const maxSize = 10 * 1024 * 1024; // 10MB in bytes
    if (file.size > maxSize) {
      return NextResponse.json(
        { error: 'File size too large. Maximum size is 10MB' },
        { status: 400 }
      );
    }

    const bytes = await file.arrayBuffer();
    const buffer = Buffer.from(bytes);
    const timestamp = Date.now();
    const randomString = Math.random().toString(36).substring(7);
    const tempFilename = `upload_${timestamp}_${randomString}.pdf`;
    const tempFilePath = join(os.tmpdir(), tempFilename);

    await writeFile(tempFilePath, buffer);

    // Extract data from PDF
    const { content: pdfContent, response: ocrResponse } = await ocrService.processFile(tempFilePath, file.name);

    // Generate script for the audio from the extracted data
    const audioScript = await openaiService.generatePodcastScript(pdfContent);

    // Generate audio from the script
    const audioFiles = await generatePodcastAudio(audioScript, process.env.OPENAI_API_KEY || '', {
      voice: 'verse',
      outputDir: audioDir,
      instructions: instructions,
      outputFormat: 'mp3',
    });

    const cleanedAudioFiles = cleanAudioPaths(audioFiles);

    // Save the data into GridDB database
    const podcastData: GridDBData = {
      id: generateRandomID(),
      audioFiles: JSON.stringify(cleanedAudioFiles),
      audioScript: JSON.stringify(audioScript),
      // @ts-ignore
      ocrResponse: JSON.stringify(ocrResponse),
    };

    const result = await dbClient.insertData({ data: podcastData });

    return NextResponse.json({
      message: 'File uploaded successfully',
      fileName: file.name,
      fileSize: file.size,
      tempFilePath: tempFilePath,
      ocrResponse: ocrResponse,
      audioFiles: audioFiles,
      audioScript: audioScript,
    });
  } catch (error) {
    console.error('Error uploading file:', error);
    return NextResponse.json(
      { error: 'Failed to upload file' },
      { status: 500 }
    );
  }
}

/** Extracted data from PDF */
const { content: pdfContent, response: ocrResponse } = await ocrService.processFile(tempFilePath, file.name);

/** Generate script for the audio from the extracted data */
const audioScript = await openaiService.generatePodcastScript(pdfContent);

/** Generate audio from the script */
const audioFiles = await generatePodcastAudio(audioScript, process.env.OPENAI_API_KEY || '', {
    voice: 'verse',
    outputDir: audioDir,
    instructions: instructions,
    outputFormat: 'mp3',
});

/** Save the data into GridDB database */
const podcastData: GridDBData = {
    id: generateRandomID(),
    audioFiles: JSON.stringify(cleanedAudioFiles),
    audioScript: JSON.stringify(audioScript),
    // ts ignore
    // @ts-ignore
    ocrResponse: JSON.stringify(ocrResponse)
}

const result = await dbClient.insertData({ data: podcastData });

const uploaded_pdf = await this.client.files.upload({
  file: {
    fileName: fileName,
    content: file,
  },
  purpose: "ocr",
});

const signedUrl = await this.client.files.getSignedUrl({
    fileId: uploaded_pdf.id,
});

const ocrResponse = await this.client.ocr.process({
    model: "mistral-ocr-latest",
    document: {
        type: "document_url",
        documentUrl: signedUrl.url,
 }
});

Create a 5-minute podcast episode script in a conversational style, using the content provided.\n\nInclude the following elements:\n\n- **Introduction**: Engage your audience with an intriguing opening statement related to the topic. Capture their attention immediately.\n\n- **Main Talking Points**: Develop 3-4 main sections discussing the central ideas or arguments. Use relatable examples and personal stories for better understanding. Maintain a conversational tone, as if you are speaking directly to the listener. Ensure natural transitions between sections to keep the flow.\n\n- **Conclusion**: Summarize the key takeaways in a concise manner, making sure to leave a lasting impression.\n\n- **Call to Action**: End with a clear and compelling call to action encouraging listeners to engage further or reflect on the topic.\n\n# Output Format\n\nWrite the script in a conversational and engaging narrative suitable for a podcast. Each section should integrate seamlessly with transitions, emulate a direct speaking style to engage the listener, and reinforce the message.\n\n# Examples\n\n**Introduction**: \"Welcome to [Podcast Name]. Today, we're diving into [Topic]. Have you ever wondered...?\"\n\n**Main Talking Points**:\n\n1. \"Let's start with [Main Idea]. It's like when...\"\n2. \"Moving on to [Next Idea], consider how...\"\n3. \"Finally, when we talk about [Final Idea], there's a story about...\"\n\n**Conclusion**: \"So, as we've learned today, [Key Takeaway 1], [Key Takeaway 2]...\"\n\n**Call to Action**: \"Think about how you can [Action]. Join us next time when we explore...\"\n\n# Notes\n\n- The script should be written to cater both to novices and those with some prior knowledge.\n- Ensure it resonates intellectually and stimulates curiosity among listeners.\n- Use transition words to guide listeners smoothly from one idea to the next.

{
  "introduction": "Welcome to our podcast! Today, we're exploring how AI can revolutionize the way we consume content by transforming PDFs into engaging audio podcasts. Have you ever wished you could listen to your documents instead of reading them? Let's dive in!",
  "main_talking_points": [
 {
      "title": "The Challenges of Manual PDF-to-Podcast Conversion",
      "content": "Manually converting PDFs into podcasts is a tedious process. It involves extracting text, summarizing complex content, and recording audio—all of which take significant time and effort. AI simplifies this by automating these steps, saving you hours of work."
 },
 {
      "title": "How AI Simplifies the Process",
      "content": "AI tools like Mistral OCR and OpenAI TTS streamline the workflow. Mistral OCR extracts text from PDFs with high accuracy, while OpenAI's models summarize and convert the text into natural-sounding audio. This ensures a seamless and efficient process."
 },
 {
      "title": "The Role of GridDB in Managing Data",
      "content": "GridDB Cloud acts as a robust storage solution for parsed text and audio files. It ensures that your data is organized, easily retrievable, and ready for future use, making the entire system scalable and efficient."
 }
 ],
  "conclusion": "In summary, AI-powered tools are transforming the way we interact with content. By automating the conversion of PDFs into podcasts, we save time, enhance accessibility, and create a more engaging learning experience.",
  "call_to_action": "Think about how you can leverage this technology in your own projects. Visit our GitHub repository to get started, and don't forget to share your feedback!"
}

  const simpleKeys = ['introduction', 'conclusion', 'call_to_action'] as const;

  for (const key of simpleKeys) {
    try {
      const text = podcastData[key];
      const fileName = `${key}.${outputFormat}`;
      const speechFile = path.join(outputDir, fileName);

      const response = await openai.audio.speech.create({
        model,
        voice,
        instructions,
        input: text
 });

      const buffer = Buffer.from(await response.arrayBuffer());
      fs.writeFileSync(speechFile, buffer);

      audioFiles[key] = speechFile;
 } catch (error) {
      console.error(`Error processing ${key}:`, error);
      throw error;
 }
 }

  // Process main talking points separately
  if (Array.isArray(podcastData.main_talking_points)) {
    for (let i = 0; i < podcastData.main_talking_points.length; i++) {
      try {
        const point = podcastData.main_talking_points[i];
        const text = point.content;
        const fileName = `talking_point_${i}.${outputFormat}`;
        const speechFile = path.join(outputDir, fileName);

        const response = await openai.audio.speech.create({
          model,
          voice,
          instructions,
          input: text
 });

        const buffer = Buffer.from(await response.arrayBuffer());
        fs.writeFileSync(speechFile, buffer);

        audioFiles[`talking_point_${i}`] = speechFile;
 } catch (error) {
        console.error(`Error processing talking point ${i}:`, error);
        throw error;
 }
 }

export interface GridDBData {
  id: string | number;
  ocrResponse: Blob;
  audioScript: string;
  audioFiles: string;
}

async function insertData({
  data,
  containerName = 'podcasts',
}: {
  data: GridDBData;
  containerName?: string;
}): Promise {
  console.log(data);
  try {
    const row = [
      parseInt(data.id.toString(), 10),
      data.ocrResponse,
      data.audioScript,
      data.audioFiles,
 ];

    const path = `/containers/${containerName}/rows`;
    return await makeRequest(path, [row], 'PUT');
 } catch (error) {
    if (error instanceof GridDBError) {
      throw error;
 }
    const errorMessage = error instanceof Error ? error.message : 'Unknown error';
    throw new GridDBError(`Failed to insert data: ${errorMessage}`, undefined, undefined, error);
 }
}