Chatbot Interface and API

This repository contains a Flask-based chatbot with a web interface and API. The application supports conversational interactions, text-to-speech (TTS), and dynamic reloading of configurations, such as model settings and embedding modifications.

Table of Contents

• Features
• Installation
• Usage
• API Endpoints
  • POST /chat
  • GET /get_config
  • POST /update_config
• Configuration
• Run the Application

Features

• Conversational chatbot with text-to-speech support.
• Dynamic configuration management with real-time reloading.
• Supports embedding-based search using K-Nearest Neighbors (KNN) with a user-defined boost value.
• Simple HTML-based frontend for chatting with the assistant.
• Speech recognition for inputs.

Installation

1. Clone the repository:

   git clone https://github.com/zackrack/adaptivechatbot
   cd adaptivechatbot
   

2. (Recommended) Use a Python virtual environment:

   python3 -m venv venv
   source venv/bin/activate
   

3. Install the required Python packages:

   pip install -r requirements.txt
   

4. Set up environment variables:

   • Make sure to set a secure Flask secret key.
   • Optionally, configure CUDA_VISIBLE_DEVICES if using a machine with multiple GPUs to only use one GPU:

     export CUDA_VISIBLE_DEVICES=0
     

5. (Optional, Configure your models and embeddings in config.json.

Usage

1. Start the Flask application:

   python app.py
   

2. Access the chatbot:

   • Visit http://localhost:55556/ in your web browser to chat with the assistant via the web interface.

3. Interact via API:

   • You can also interact with the chatbot and update its configuration via the provided API endpoints.

API Endpoints

POST /chat

This endpoint processes user input and returns a chatbot response, optionally generating an audio response using TTS.

• URL: /chat
• Method: POST
• Request Format: JSON or Form Data
• Request Parameters:
  • message (string): The user's input to the chatbot.
• Response:
  • message (string): The assistant's response.
  • audio_url (string): A URL to the generated audio file (if TTS is enabled).

Example Request (JSON):

curl -X POST http://localhost:55556/chat -H "Content-Type: application/json" -d '{"message": "Hello!"}'

Common Error Solutions

1. Ensure Praat is Installed

The Praat application must be installed on your system. Update the praat_executable path in audio.py accordingly. Depending on your situation, you may need to build Praat from source.

Steps to Build Praat from Source (conda):

Note that you should install praat in a separate location that is not this repository since the folder holding the scripts here is also named 'praat.' Alternatively, you can rename the folder in this repository.

1. Clone the Praat repository:

   git clone https://github.com/praat/praat.git
   

2. Install the required dependencies:

   conda install cmake gcc gxx gtk3 alsa-lib -c conda-forge
   

3. Navigate to the cloned repository:

   cd praat
   

4. Create a build directory:

   mkdir build
   cd build
   

5. Configure the build:

   cmake ..
   

6. Compile the source code:

   make -j$(nproc)
   

2. Ensure FFmpeg is Installed

Make sure FFmpeg is installed on your system. You can install it using apt (Linux), brew (macOS), or from the FFmpeg official website.

Todo:

• Dynamically delete generated audio after next message (right now it's staying in the cache)
• Implement download and inference with random forest classifier.
• Separate word lists into levels
• Retrieve words based on user level
• Train English TTS Model for Style-BERT-VITS2