Trigger Word Assistant

A voice based assistant application which executes the assigned command on detecting the trigger word from the user voice. A trigger word is a word to which the model gets triggered once it detects it. Like for example “Hey Cortana” or “Ok Google”. Here the trigger word used is “activate”.

The user needs to speak through the microphone , then the audio data is processed and fed to the model. The model after detecting the trigger word does the action specified in the code, like I coded it to open Chrome everytime it detects that the user has spoken activate.

The Model used for detecting trigger word uses a deep Recurrent Neural Network with Gated Recurrent Units (GRU). The model is build using Keras.

After the audio is processed by the model a notification alert is inserted at the timesteps where the model detected the trigger word in the audio file, you can listen to it inside the output folder.

Inside the audio folder the recorded voice of user is stored.

There are 2 versions for this application:

1. Python file incase you want to directly see what is happening.

Just run the script and wait for the message “Started Listening” , then speak through the microphone of the Computer and see it work.
Inside the output folder the spectogram of the processed audio is stored. Here is a gif showing this in action

Note: The script continues to run till the desired application opened by the command after trigger detection is not closed.

2. Jupyter notebook explaining everything.

Detailed walkthrough of the code including option to listen to the actual output processed by the model.

Credits:

This project is based on the assignment from Sequence Models Specialization by Deeplearning.ai on Coursera. https://www.coursera.org/learn/nlp-sequence-models/home/welcome