Skip to content

ffgiardina/wake

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Wake word detection

This notebook implements a wake-word detector. A wake word detector is used to trigger an event in a program upon the utterance of a wake word (similar to "Hey Siri" in Apple devices and "OK Google" in Google devices). It uses a deep recurrent neural network based on gated recurrent units. The training and development data is synthesized from audio samples (of backgrounds, wake words, and negatives), which can easily be generated using the recording section in this notebook. The pre-trained weights are then loaded and the model is fine tuned to the newly added audio samples.

This notebook uses the model and pre-trained weights from the Coursera course on sequence models.

As an example see the following figure. It shows the spectrogram of a sequence of negative words (Green, Purple, Yellow) and the wake word (Activate), and the model's prediction of when the wake word is said (for signals above the red horizontal line).

About

Jupyter notebook running and training a simple wake-word detector based on a deep recurrent neural network.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages