shekhuverma/TalkGPT

An SDK to integrate openAI and Deepgram API to convert chat bots into voice bots.

TalkGPT

Was not able to think any other cool name :D

Table of Contents

• TalkGPT
  • Table of Contents
  • Installation
  • Usage
  • Testing
  • Rough Implementation Idea
  • Assumptions Made
  • Scope for Improvements/ Known issues

Installation

1. Clone the repository:

   git clone https://github.com/shekhuverma/TalkGPT.git
   cd TalkGPT

2. Create a virtual environment and activate it

   python -m venv venv

   Activate it using

   #on windows
   venv\Scripts\Activate
   #on linux
   source venv/Scripts/Activate

3. Install dependencies:

   pip install -r requirements.txt

Usage

1. Configure the environment variables in the template.env file and rename it to .env

2. Run the CLI application:

   python cli.py

3. Optionally you can set pre-commit hooks using

   pre-commit install
   

Testing

All the tests are done using Pytest.

To run all the tests, run the following command

pytest tests

or you can use tox for testing and linting

tox

This will run all the tests, formats and lints the code using Ruff.

Rough Implementation Idea

1. Get the data from Pyadio Input stream using callback -> pass the data to Deepgram and save the response in a queue.
2. Get the data from queue in another async loop and for each item in queue pass the data to LLM and return the LLM result into OpenAI speech to text.

Assumptions Made

1. The OpenAI text to speech was only working best for the following settings. sample_rate=22050, chunk_size=2048

Scope for Improvements/ Known issues

• Not able to exit cleanly at times.
• Add more test cases.
• Some performance metrics are not working as expected. Will try to fix them
• Seprate the user and dev dependencies (By using PDM or poetry)

Note - Tested and developed on Windows machine on python version 3.12