make your own siri

this is a trigger word detection project made with python and tensorflow and pytorch so the project can be divided into phases so let me walk you through it

create dataset

so we need two categories of audio clips [positive , negative ] or ones have the trigger word and ones are not so the first python script we wrote is recording 20 clips for each category and gave them unique ids and save them as wav files

---

positive clips : just record the word you want to use it as trigger word

---

negative clips : record any kind of words or just let the record empty and it better to do both

adding noise

so we need to add some kind of background noise to the records to make the model learn better and do some kind of data augemention

train the model

you need to make a data pipeline to process the data after that you create the model and train it

inference

finially you want to make live prediction

how to run the app

• clone the repo

git clone the repo 
pip install -t requirments.txt 

• create the dataset

python scripts/recording.py 

• adding noise

python scripts/adding_noise.py

• the model

tensorflow => open the tensorflow_trigger_word.ipynb notebook and run all cells to train the model and save it

---

pytorch => open the pytorch_trigger_word.ipynb notebook and run all cells to train the model and save it

• do inference if your followed tensorflow notebook

python scripts/tensorflow_inference.py 

---

if you followed pytorch notebook

python scripts/pytorch_inference.py