Wake Word Detection

Team Members:

Yilei Fu: yf20@rice.edu

Yongze Yin: yy70@rice.edu

Zhen Cao: zc36@rice.edu

Xiru Huang: xh29@rice.edu

Program Structure:

A continuous loop was applied on this task, for constantly detecting the voice input several times per second.

A microphone on Arduino was responsible for audio data collection.

After audio data was collected, the program send the data into Feature Provider, to provide spectrogram of the data.

A pre-trained model was applied on the spectrogram multiple times per second, to decide the word is "yes" or "no" or other words and output a command. According to the command, the LED will shine on different colors. If there are multiple words detected in the same period(3 secs), the LED will output combined color..