Skip to content

mohammedyunus009/dnn_acoustic_rec

Repository files navigation

Introduction (machine learning ,sound recognition)

This is a keras implementation project This project is used to recognize environmental sounds ,with a high accuracy of 80% percent ,it has a deep neural network (LSTM can be implemented),can be used to train on the extracted features of audio files in https://github.com/mohammedyunus009/dev_datasets and https://github.com/mohammedyunus009/datasets.

It has the capability to recognize on 15 diffrent classes of sounds such as :

'bus' 'cafe/restaurant' 'car' 'city_center' 'forest_path'
'grocery_store' 'home' 'beach' 'library' 'metro_station'
'office' 'residential_area' 'train' 'tram' 'park'

vtu under graduate project (visvervaraya university of technology)

Requirements

This library runs with keras.

pip install -r requirements.txt

OR for conda and linux users run sh setup.sh

Quickstart

STEP 1 configure the config file in src folder

STEP 2

*python calculate_logmel.py to extract features and pickle in memory

STEP 3

  • python kera_model.py --dev_train to run in development mode (train on development dataset)

  • python kera_model.py --eva_train to run in evaluation mode (train on evaluation dataset) Reconfigure the configuration file in src

  • python kera_model.py --dev_recognize used to calculate the accuracy of the model in development mode

  • python kera_model.py --eva_recognize used to calculate the accuracy of the model in evaluation mode

STEP 4

  • python session.py used to put in production and test new files

Contact me for more information