https://github.com/pragativerma18/call-analysis
This project analyzes call data to measure overtalk and silence metrics while also detecting profanity and sensitive information using regex-based and LLM-based approaches.
https://github.com/pragativerma18/call-analysis
chatgpt conversational-agents llm python regex streamlit streamlit-webapp
Last synced: 4 months ago
JSON representation
This project analyzes call data to measure overtalk and silence metrics while also detecting profanity and sensitive information using regex-based and LLM-based approaches.
- Host: GitHub
- URL: https://github.com/pragativerma18/call-analysis
- Owner: PragatiVerma18
- Created: 2025-03-28T17:37:36.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2025-03-29T07:51:27.000Z (7 months ago)
- Last Synced: 2025-03-29T08:22:01.300Z (7 months ago)
- Topics: chatgpt, conversational-agents, llm, python, regex, streamlit, streamlit-webapp
- Language: Python
- Homepage: https://convo-insight.streamlit.app/
- Size: 15.6 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Call Analysis and Profanity Detection
> This project analyzes call data to measure overtalk and silence metrics while also detecting profanity and sensitive information using regex-based and LLM-based approaches.
## Features
- **Call Analysis**: Computes silence and overtalk percentages.
- **Visualization**: Generates charts for call metrics.
- **Profanity Detection**: Uses regex and ChatGPT API for comparison.
- **Sensitive Data Detection**: Identifies sensitive information like account numbers and SSNs.
- **Streamlit App**: Provides an interactive UI for users to analyze call data.
> Find detailed technical report [here](./technical-report.md)
## Setup Instructions
### Prerequisites
Ensure that the following dependencies are installed:
- Python 3.8+
- Virtual environment
### Installation
1. Clone the repository:
```shell
https://github.com/PragatiVerma18/call-analysis.git
cd call-analysis
```
2. Create and activate a virtual environment:
```shell
python -m venv venv
source venv/bin/activate
```
3. Install dependencies:
```shell
pip install -r requirements.txt
```
4. Set up environment variables in `.env`:
```shell
OPENAI_API_KEY="your-api-key"
```
5. Run the Streamlit App:
```shell
streamlit run app.py
```
> This will launch a local server where users can upload call data and visualize metrics.
## Streamlit App Overview
Once the app is running, you can interact with the following features:
### 1. File Upload:
- Upload a `.json` or `.yaml` file containing call conversation data. Find example file here - [example-conversation.json](./example-conversation.json).
- The app will parse the file and extract relevant information such as speaker activity, timestamps, and text.
### 2. Tabs for Analysis and Visualization:
- The app is divided into two main tabs:
- **Analysis Results**:
- Select an analysis method (`Regex` or `LLM`) to detect profanity or privacy violations.
- View flagged utterances in the conversation.
- Get feedback on whether any issues were detected.
- **Call Quality Metrics**:
- Choose from multiple visualization options to analyze call quality:
- **Pie Chart**: Displays the proportions of silence, overtalk, and normal conversation.
- **Bar Chart**: A horizontal bar chart (Gantt Chart style) for comparing silence, overtalk, and normal conversation percentages.
- **Dual Line Chart**: Plots the speaking activity of both the agent and the customer over time, highlighting overtalk segments.
### 3. Analysis Features:
- **Profanity Detection**:
- Detects offensive language in the conversation using either regex-based or LLM-based methods.
- **Privacy Violation Detection**:
- Identifies sensitive information such as account numbers, SSNs, and other personal identifiers.
### 4. Call Quality Metrics:
- **Silence Percentage**:
- Measures the proportion of the call where neither party is speaking.
- **Overtalk Percentage**:
- Measures the proportion of the call where both parties are speaking simultaneously.
- **Normal Conversation**:
- Represents the remaining portion of the call where only one party is speaking at a time.
### 5. Interactive Visualizations:
- The app provides intuitive visualizations to help users understand call dynamics:
- **Pie Chart**: For an overall breakdown of silence, overtalk, and normal conversation.
- **Bar Chart**: For comparing metrics side by side.
- **Dual Line Chart**: For analyzing speaker activity over time.
### 6. Insights:
- Based on the analysis, the app provides insights into call quality and compliance.
## Demo
> Find detailed technical report [here](./technical-report.md)