marcboeker/journal-gpt

If Alexa and ChatGPT had a child. Prototype of using OpenAI Whisper and GPT through a web interface.

JournalGPT

This is just a proof of concept (PoC), not a fully developed product. However, it serves as an idea for combining text-to-speech and LLMs (Large Language Models).

Are you tired of writing long journal entries every day? Are you also tired of reading them in the future? Let's simplify things by using text-to-speech technology to dictate your daily experiences. The LLM (GPT-3.5 Turbo) will create a summary and categorize it for you in these categories:

• Your day
• Learnings
• To-dos

Prerequisites

Before running this project, you need to have the following installed on your system:

• Python 3
• pip

Installation

To install the project dependencies, run the following command:

pip install -r requirements.txt

Usage

OPENAI_API_KEY=<your-api-key> python app.py

This will start a web server at http://localhost:5000. Click on Record to begin dictating your daily experiences. Once you're done, click 'Stop' and wait for the input to be processed.

Mobile Usage

To use it in Safari on iOS, you need to access it via HTTP_S_. The simplest way is to use ngrok. Simply install and sign up for ngrok, then run ngrok http 5000.

Text to Speech

We're utilizing the Whisper ASR (Automatic Speech Recognition) system developed by OpenAI. You have two options: running it locally by setting the environment variable USE_LOCAL_WHISPER=1 or using OpenAI's Whisper API, which is the default and faster. When running locally, the large model (approximately 2-3 GB) will be downloaded and the TTS will be processed on your machine.