An NLP based Chatbot over a simple fully connected neural network architecture using Tensorflow and tflearn. Trained over a custom dataset specified in the JSON file.
Run pip install -r requirements.txt in your terminal to install all required libraries
The dataset.json contains the intents on which the model is trained. Each pattern (sentence) and response is given a particular tag. The model classifies each input sentence under a tag and gives out a random answer corresponding to that tag. The json file can be formatted according to the user's requirements.
In Python versions above 3.6 PyAudio is not a supported library and installing PyAudio directly using pip install pyaudio fails and needs to be installed separately.
-
Windows
For that, individually install the wheel file given (this is for Python 3.7, find suitable wheel files here) writing the following line in your terminal.
pip install <.wh file name> -
Ubuntu/Linux
$ sudo apt-get install python3-pyaudio
-
Run
trainModel.pyto train the Fully Connected Network on the dataset. You can change the number of epochs or layers accordingly, the current architecture gave good results with a ~95% accuracy on predicting tags. -
Run
textChatbot.pyfor text based chatbot with GUI incorporated -
Run
voiceBot.pyfor voice recognition based chatbot
