Interrogate and Summarize Speech Audio on Local Machine

About

This repo contains a Streamlit app. The Streamlit app allows dragging and dropping WAV/MP3 files of audio. It converts the speech to text using Whisper, and then summarizes the text into meeting notes using Llama2. Once the audio has been transcribed, the user can input their own prompts to ask questions and generate content. This application runs entirely on local machine, meaning it can be used to summarize confidential meetings without concern of data exposure.

Dependencies

• python 3.7 or above (not in requirements.txt)
• Ollama

Instructions

1. Ensure python is installed
2. Ensure Git is installed
3. Install Ollama by visiting https://ollama.com/download
4. Ensure Ollama is in your Applications directory
5. Manually launch Ollama and continue through prompts until you click "Finish"
6. Clone this repository to your intended working directory
7. In Terminal, "ollama pull llama2"
8. In Terminal, navigate to your working directory
9. In Terminal, "pip install -r requirements.txt"
10. In Terminal, "streamlit run audio_summary_app.py"

Limitations

This application is designed to run on a local machine, therefore running large models locally requires sufficient CPU/GPU and memory resources.