CrispenGari/animal-sound-classification

this is a simple artificial neural network model using deep learning and torch-audio to classify cats and dog sounds.

Animal Sound Classification (Cats Vrs Dogs Audio Sentiment Classification)

This is a simple audio classification api build to classify the sound of an audio, weather it is the cat or dog sound.

Response

Given a .wav audio the model will classify what does the sound the audio belongs to either cat or dog.

{
  "predictions": {
    "class": "dog",
    "label": 1,
    "probability": 1.0
  },
  "success": true
}

Starting the server

To start server and start audio classification first you need to make sure you are in the server folder and run the following commands:

1. creating a virtual environment

virtualenv venv && .\venv\Scripts\activate.bat

2. installing packages

pip install -r requirements.txt

3. Starting the server

python api/app.py

The server will start on a default port of 3001 and you will be able to make api request to the server to do audio classification.

Model Metrics

The following table shows all the metrics summary we get after training the model for few 15 epochs.

model name	model description	test accuracy	validation accuracy	train accuracy	test loss	validation loss	train loss
cats-dogs-sound-cnn.pt	audio sentiment classification for dogs and cats CNN.	90.7%	90.7%	93.5%	0.621	0.218	0.209

Classification report

The following is the classification report for the model on the test dataset.

#	precision	recall	f1-score	support
accuracy	-	-	90%	2305
macro avg	91%	90%	90%	2305
weighted avg	92%	89%	90%	2305

Confusion matrix

The following figure shows a confusion matrix for the classification model.

Audio Sentiment classification

If you hit the server at http://localhost:3001/classify you will be able to get the following expected response that is if the request method is POST and you provide the file expected by the server.

Expected Response

The expected response at http://localhost:3001/classify with a file audio of the right format will yield the following json response to the client.

{
  "predictions": {
    "class": "dog",
    "label": 1,
    "probability": 1.0
  },
  "success": true
}

Using curl

Make sure that you have the audio named cat.wav in the current folder that you are running your cmd otherwise you have to provide an absolute or relative path to the audio.

To make a curl POST request at http://localhost:3001/classify with the file cat.wav we run the following command.

# for cat
curl -X POST -F audio=@cat.wav http://127.0.0.1:3001/classify

# for dog
curl -X POST -F audio=@dog.wav http://127.0.0.1:3001/classify

Using Postman client

To make this request with postman we do it as follows:

1. Change the request method to POST at http://127.0.0.1:3001/classify
2. Click on form-data
3. Select type to be file on the KEY attribute
4. For the KEY type audio and select the audio you want to predict under value
5. Click send

If everything went well you will get the following response depending on the audio you have selected:

{
  "predictions": { "class": "dog", "label": 1, "probability": 1.0 },
  "success": true
}

Using JavaScript fetch api.

1. First you need to get the input from html
2. Create a formData object
3. make a POST requests

const input = document.getElementById("input").files[0];
let formData = new FormData();
formData.append("audio", input);
fetch("http://127.0.0.1:3001/classify", {
  method: "POST",
  body: formData,
})
  .then((res) => res.json())
  .then((data) => console.log(data));

If everything went well you will be able to get expected response.

{
  "predictions": { "class": "dog", "label": 1, "probability": 1.0 },
  "success": true
}

Notebooks

• All notebooks for training and saving the models are found in the notebooks folder of this repository.