In this project is presented a simple method to train an MLP neural network for audio signals. The trained model can be exported on a Raspberry Pi (2 or superior suggested) to classify audio signal registered with USB microphone

License: MIT License

Python 100.00%

machine-learning multilayer-perceptron-network tensorflow tensorflow-models audio-analysis librosa raspberry dataset audio-signals raspberry-pi sound-classification

sound-classification-on-raspberry-pi-with-tensorflow's Introduction

SOUND CLASSIFICATION WITH TENSORFLOW ON RASPBERRY PI

BUILD THE PROJECT

The project is developed and tested with Python 2.7.

Install following Python libraries on your PC/Workstation and Raspberry Pi:

Tensorflow, Scikit-learn, Librosa

Install following library on your Raspberry only:

Sounddevice

DOWNLOAD UrbanSound8K DATASET

https://serv.cusp.nyu.edu/projects/urbansounddataset/urbansound8k.html

TRAIN THE MODEL

Set the right path where you downloaded the dataset in your code.

Set the right path where you want to save the trained model.

Run "trainModel.py" on your PC/Workstation.

RUN THE MODEL

Export the trained model on you Raspberry Pi ('model.meta', 'model.index', 'checkpoint', 'model.data-00000-of-00001').

Export 'fit_params.npy' on your Raspberry Pi.

Run "classiPi.py" on your Raspberry and enjoy!

REMEMBER TO

Remember to reference this project in your works.

AUTHORS

Gianluca Paolocci, University of Naples Parthenope, Science and Techonlogies Departement, Ms.c Applied Computer Science https://www.linkedin.com/in/gianluca-paolocci-a19678b6/

Luigi Russo, University of Naples Parthenope, Science and Techonlogies Departement, Ms.c Applied Computer Science

CONTACTS

if you have problems, questions, ideas or suggestions, please contact me to:

[email protected]

sound-classification-on-raspberry-pi-with-tensorflow's People

Contributors

Stargazers

Watchers

sound-classification-on-raspberry-pi-with-tensorflow's Issues

Parent directory does not exist on model save

After epochs done, it throws an error on Linux machine with TF 1.11.0

I fixed it on this PR #2

Bird classification

Hello,

Is it possible to only classify birds or create a response when a bird call is heard?

Or is this beyond the scope of this project?

Kind regards,

Error lines 38

The system said "did you mean print(sub_dir)?
How can I fix it??

I had an error on training in trainModel.py

I had an error on this line
train_x, test_x, train_y, test_y = train_test_split(features, labels, test_size=0.3, random_state=0)

they said
ValueError: With n_samples=0, test_size=0.3 and train_size=None, the resulting train set will be empty. Adjust any of the aforementioned parameters.

How can I solve this?

Cost is not defined

First of all thank you for sharing your code!
Do you know how I can fix this issue?
https://i.gyazo.com/28530cdfc1ae2728baa2111751bf3511.png

Thank you in advance!

Which Librosa version did you used?

UnboundLocalError: local variable 'features' referenced before assignment

At classiPi.py
I can`t find features is assigned.
how can i solve this problem?

how can I add new class

hi
How can I add new classes to the code?
beside classiPi.py doesn't work with me. it always as to install liborsa where there is no liborsa for raspian.
where can I find a dataset that is appropriate to the code?
thanks so much.
your code is very helpful.

How to use it with your own dataset?

are fsID and salience necessary and if so, what do they mean?
where to set the filename of the csv in the metadata subfolder?
I get ValueError: With n_samples=0, test_size=0.3 and train_size=None, the resulting train set will be empty. Adjust any of the aforementioned parameters. with my training set. What is the likely cause? "Not found"?

gianlucapaolocci / sound-classification-on-raspberry-pi-with-tensorflow Goto Github PK