Project LATACA: Local Automated Transcription and Content Analysis

Overview

Project LATACA is designed to facilitate automatic transcription of interviews and to apply coding schemes for sentiment analysis. It features a front-end interface built with Streamlit and leverages machine learning models for speech recognition and content analysis.

Installation

Prerequisites:

Python 3.9 or higher
Conda package manager

Setup Instructions:

Python Installation: Obtain Python from the official Python website.
Conda Installation: Install Conda by following the official Conda installation guide.
Fork the Repository: Fork the project repository from GitHub to your account.
Clone Your Fork: Clone the repository to your local machine. git clone https://github.com/YOUR_USERNAME/LATACA.git or use the GitHub Desktop client.
Conda Environment:
- Navigate to the project directory:
```
cd ***PATH_TO_YOUR_CLONE***
```
- Create a conda environment using the environment.yaml file provided in the repository:
```
conda env create -f environment.yaml
```
- Activate the conda environment:
```
conda activate transcribe
```
- Load the ASR model from the Internet:
```
pip install git+https://github.com/akashmjn/tinydiarize.git
```
Run the Streamlit App:
- Execute the following command:
```
streamlit run app.py
```

Models Used

Whisper: An automatic speech recognition model for transcribing the interviews. We use a finetuned version that supports speaker diarization (recognising the differnt speakers in the audio) for 2 people. This model is based on the small.en model.
BART: A transformer model used for sequence classification tasks in the content analysis.

Usage

Launch the Streamlit app and select your interview audio file.
The app will transcribe the audio and then apply the specified coding schemes.
Adjust the number of sentences for analysis and confidence levels for classification as needed.
Results can be exported in JSON or DOCX formats, and further analysis can be conducted with the provided content analysis tools.
Content Analysis Configuration

After uploading your interview file, you can:

Set the unit of analysis for the ML model (number of sentences).
Modify codes for content analysis by entering different labels.
Adjust confidence levels to fine-tune the ML model's code assignment.

Data Processing and Exporting

Once the interview is uploaded:

The app transcribes the content using the Whisper model.
Users can view the transcription and apply a coding scheme for sentiment analysis.
The processed data can be exported in JSON, CSV, or DOCX format.

Code Customization

Users have the flexibility to:

Edit, add, or remove codes for content analysis.
View the resultant codes and their confidence scores, and make adjustments as necessary.

Language Support

Currently, LATACA supports English transcription and analysis.
Development for other language support is underway.

Features

Transcription using Whisper model.
Sentiment analysis and content coding using BART Large model.
Data visualization with Seaborn and Plotly.
Results export for further analysis in tools like NVivo.

Contributing

For feature requests, please open an issue in the GitHub repository.
To contribute to the project, fork the repo, make changes, and open a pull request for review.

Contact

For inquiries or assistance, reach out via LinkedIn, GitHub, or Email.

License

This codebase is free to use and modify for non-commercial use. Selling any of the technology used here to further parties requires explicit written consent.

Note: The accuracy of the transcription and content analysis may vary based on audio quality and the specificity of the coding schemes. It is recommended to review the transcriptions and analysis results for accuracy.

Made with 🧡 using Streamlit.

JonasWeinert/LATACA