Local-NotebookLM

NEW! A stand-alone app has been released link.

A local AI-powered tool that converts PDF documents into engaging audio—such as podcasts or custom audio content—using local LLMs and TTS models.

Features

• PDF text extraction and processing
• Customizable audio generation (podcasts, summaries, interviews, and more) with different styles and lengths
• Support for various LLM providers (OpenAI, Groq, LMStudio, Ollama, Azure)
• Text-to-Speech conversion with voice selection
• Flexible pipeline with many options for content, style, and voices
• Programmatic API for integration in other projects
• FastAPI server for web-based access
• Example podcast included for demonstration

Here are quick examples, can you guess what paper they're talking about?

Your browser does not support the audio element. You can manually listen/download here: <a href="https://raw.githubusercontent.com/Goekdeniz-Guelmez/Local-NotebookLM/main/examples/podcast_example_casual.wav">Casual example</a>.

Your browser does not support the audio element. You can manually listen/download here: <a href="https://raw.githubusercontent.com/Goekdeniz-Guelmez/Local-NotebookLM/main/examples/podcast_example_genz.wav">Gen-Z example</a>.

If your browser still blocks embedded playback on GitHub, use direct links: - Casual example (.wav) - Gen-Z example (.wav)

Prerequisites

• Python 3.9+
• Local LLM server (optional, for local inference)
• Local TTS server (optional, for local audio generation)
• At least 8GB RAM (32GB+ recommended for local models)
• 10GB+ free disk space

Installation

From PyPI

pip install local-notebooklm

From source

1. Clone the repository:

git clone https://github.com/Goekdeniz-Guelmez/Local-NotebookLM.git
cd Local-NotebookLM

1. Create and activate a virtual environment (conda works too):

python -m venv venv
source venv/bin/activate  # On Windows, use: venv\Scripts\activate

1. Install the required packages:

pip install -r requirements.txt

Running with Docker

You can run Local-NotebookLM using Docker for both the Web UI and API modes.

Prerequisites

• Docker installed on your system

Steps

1. Build the Docker image:

   bash docker build -t local-notebooklm-ui .

2. Run the Gradio Web UI:

   bash docker run -p 7860:7860 local-notebooklm-ui The Web UI will be available at http://localhost:7860.

3. Run the FastAPI API server:

   bash docker run -e APP_MODE=api -p 8000:8000 local-notebooklm-ui The API server will be available at http://localhost:8000.

Optional pre requisites

Local TTS server

• Follow one installation type (docker, docker-compose, uv) at https://github.com/remsky/Kokoro-FastAPI
• Test in your browser that http://localhost:8880/v1 return the json: {"detail":"Not Found"}

Example Output

The repository includes an example podcast in examples/podcast.wav to demonstrate the quality and format of the output. The models used are: gpt4o and Mini with tts-hs on Azure. You can listen to this example to get a sense of what Local-NotebookLM can produce before running it on your own PDFs.

Usage

Command Line Interface

Run the script with the following command:

python -m local_notebooklm.make_audio --pdf PATH_TO_PDF [options]

Available Options

Option	Description	Default
--pdf	Path to the PDF file (required)	-
--output_dir	Directory to store output files	./output
--llm_model	Ollama LLM model name	gemini-3-flash-preview:cloud
--language	Language for the audio output	english
--format_type	Output format type (summary, podcast, article, interview, panel-discussion, debate, narration, storytelling, explainer, lecture, tutorial, q-and-a, news-report, executive-brief, meeting, analysis)	podcast
--style	Content style (normal, casual, formal, technical, academic, friendly, gen-z, funny)	normal
--length	Content length (short, medium, long, very-long)	medium
--is-vlm	Enable vision mode so extracted PDF images are also sent to the LLM	False
--num_speakers	Number of speakers in audio (1, 2, 3, 4, 5)	2 (for podcast/interview)
--custom_preferences	Additional focus preferences or instructions	None

Format Types

Local-NotebookLM supports both single-speaker and multi-speaker formats:

Single-Speaker Formats: - summary - narration - storytelling - explainer - lecture - tutorial - news-report - executive-brief - analysis

Two-Speaker Formats: - podcast - interview - panel-discussion - debate - q-and-a - meeting

Multi-Speaker Formats: - panel-discussion (3, 4, or 5 speakers) - debate (3, 4, or 5 speakers)

Example Commands

Basic usage:

python -m local_notebooklm.make_audio --pdf documents/research_paper.pdf

Customized podcast:

python -m local_notebooklm.make_audio --pdf documents/research_paper.pdf --format_type podcast --length long --style casual

With custom preferences:

python -m local_notebooklm.make_audio --pdf documents/research_paper.pdf --custom_preferences "Focus on practical applications and real-world examples"

Specify number of speakers:

python -m local_notebooklm.make_audio --pdf documents/research_paper.pdf --format_type panel-discussion --num_speakers 3

Enable multimodal transcript generation (text + PDF images):

python -m local_notebooklm.make_audio --pdf documents/research_paper.pdf --is-vlm

Programmatic API

You can also use Local-NotebookLM programmatically in your Python code:

from local_notebooklm.processor import generate_audio

generate_audio(
    pdf_path="documents/research_paper.pdf",
    output_dir="./test_output",
    llm_model="qwen3:30b-a3b-instruct-2507-q4_K_M",
    language="english",
    format_type="interview",
    style="professional",
    length="long",
    num_speakers=2,
    custom_preferences="Focus on the key technical aspects"
)

Gradio Web UI

Local-NotebookLM now includes a user-friendly Gradio web interface that makes it easy to use the tool without command line knowledge:

python -m local_notebooklm.web_ui

By default, the web UI runs locally on http://localhost:7860. You can access it from your browser.

Web UI Screenshots

The main interface of the Local-NotebookLM web UI

Web UI Options

Option	Description	Default
--share	Make the UI accessible over the network	False
--port	Specify a custom port	7860

Example Commands

Basic local usage:

python -m local_notebooklm.web_ui

Share with others on your network:

python -m local_notebooklm.web_ui --share

Use a custom port:

python -m local_notebooklm.web_ui --port 8080

The web interface provides all the same options as the command line tool in an intuitive UI, making it easier for non-technical users to generate audio content from PDFs.

FastAPI Server

Start the FastAPI server to access the functionality via a web API:

 python -m local_notebooklm.server

By default, the server runs on http://localhost:8000. You can access the API documentation at http://localhost:8000/docs.

Pipeline Steps

1. PDF Extraction
2. Extracts and cleans text from the provided PDF.
3. Transcript Generation
4. Generates a transcript or script based on the extracted content and user options.
5. Audio Generation
6. Converts the optimized transcript to audio using the specified TTS model and outputs the final audio file.

Pipeline Diagram

flowchart TD
    subgraph "Main Controller"
        generate_audio["generate_audio()"]
    end

    subgraph "Pipeline Steps"
        extractPDF["PDF Extraction"]
        transcript["Transcript Generation"]
        ttsOpt["TTS Optimization"]
        audioGen["Audio Generation"]
    end

    generate_audio --> extractPDF
    extractPDF --> transcript
    transcript --> ttsOpt
    ttsOpt --> audioGen
    audioGen --> outputFile["Audio File"]

Multiple Language Support

Local-NotebookLM now supports multiple languages. You can specify the language when using the programmatic API or through the command line.

Important Note: When using a non-English language, ensure that both your selected LLM and TTS models support the desired language. Language support varies significantly between different models and providers. For optimal results, verify that your chosen models have strong capabilities in your target language before processing.

Output Files

The pipeline generates the following files:

• segments/podcast_segment_*.wav: Individual audio segments
• podcast.wav: Final concatenated podcast audio file

Troubleshooting

Common Issues

1. PDF Extraction Fails
2. Try a different PDF file
3. Check if the PDF is password-protected

4. Ensure the PDF contains extractable text (not just images)

5. LLM Call Fails

6. Ensure the LLM model exists

7. API Connection Errors

8. Verify your API keys are correct
9. Check your internet connection

10. Ensure the API endpoints are accessible

11. Out of Memory Errors

12. Reduce the chunk size in the configuration
13. Use a smaller model

14. Close other memory-intensive applications

15. Audio Quality Issues

16. Try different TTS voices
17. Adjust the sample rate in the configuration
18. Check if the TTS server is running correctly

Getting Help

If you encounter issues not covered here, please: 1. Check the logs for detailed error messages 2. Open an issue on the GitHub repository with details about your problem 3. Include the error message and steps to reproduce the issue

Requirements

• Python 3.9+
• PyPDF2
• tqdm
• numpy
• soundfile
• requests
• pathlib
• fastapi
• uvicorn

Full requirements are listed in requirements.txt.

Acknowledgments

• This project uses various open-source libraries and models
• Special thanks to the developers of LLaMA, OpenAI, and other AI models that make this possible

Best Gökdeniz Gülmez

---

---

Citing Local-NotebookLM

The Local-NotebookLM software suite was developed by Gökdeniz Gülmez. If you find Local-NotebookLM useful in your research and wish to cite it, please use the following BibTex entry:

@software{
  Local-NotebookLM,
  author = {Gökdeniz Gülmez},
  title = {{Local-NotebookLM}: A Local-NotebookLM to convert PDFs into Audio.},
  url = {https://github.com/Goekdeniz-Guelmez/Local-NotebookLM},
  version = {0.1.5},
  year = {2025},
}