Project: Marvin

Audio categorization. Deep learning

Artificial Intelligence  
Domain             : Audio recognition
Sub-Domain         : Audio categorization
Architectures      : Sequential
Application        : Voice recognition for the voice assistance

The objective is to develop audio recognition for initiating commands with brief trigger words, specifically tailored for non-commercial voice assistants. The designated keyword for recognition is "Marvin."

Description of data

The dataset for training encompasses audio snippets covering various conversational scenarios, including background noise, to enhance the model's ability to recognize "Marvin" amidst real-world conditions.

Metrics of the Final Model

Final Model             : marvin_voice.tflite
Accuracy :              : Val - 99,  Test - 98

Tools / Libraries

Languages               : Python
Tools/IDE               : Anaconda
Libraries               : Keras, TensorFlow
Virtual Environment     : pipenv

Scripts and Notebook:

Notebook                : [martvin_audio_classification.ipynb]
Predict Script          : [predict.py]
Train Script            : [train.py]
Test Script             : [test_manual.py]
Test Script             : [test.py]
Test Script             : [test_AWS.py]

Run the Model as is

Steps to run the scripts/notebooks as is:

1. Clone the repo by running the following command:

   git clone https://github.com/Justlesia/voice_model.git
   

2. Download data:

   https://drive.google.com/uc?export=download&id=11NAqQ-T2xRd1LHT3hhT7W94F15fn6dvn

Using Docker

1. Build and run the application using the commands:

   docker build -t predict .
   docker run --rm -p 9696:8080 predict
   
   

2. Open another terminal/prompt and run test_docker.py

   python test.py
   

   If the prediction services give a prediction, then it means the docker container is working.

Deployment using AWS Lambda

Skip this step and go through to the 2 step.

1. Go to use awscli in the repository and type the following commands. The lambda in the folder will be used to create a docker image and run the app.

   aws ecr get-login-password --region eu-west-1 | docker login --username AWS --password-stdin 362414735980.dkr.ecr.eu-west-1.amazonaws.com
   docker build -t marvin-predict .
   docker tag marvin-predict:latest 362414735980.dkr.ecr.eu-west-1.amazonaws.com/marvin-predict:latest
   docker push 362414735980.dkr.ecr.eu-west-1.amazonaws.com/marvin-predict:latest
   

2. Go to use setup and test lambda service:

3. Now the app running at the address https://bjh5fym1el.execute-api.eu-west-1.amazonaws.com/test/predict. To test if the docker container was built properly and running, go to the test_aws.py

   python test_AWS.py
   

   If the prediction services give prediction, the service app is working.