Description

Have you ever wanted to check the precise accuracy of your music practice based on professional samples? Before creating this program, I wanted to relearn how to play the piano but I was unsure of how correct my practice was. Letoplay is a music accuracy program that gauges the correctness of your music based on training samples based on tempo and tune. Here are some features of the program:

1. Percent Accuracy of Music
2. Training on which parts of the song need improvement (currently in development)
3. Easy to use UI (currently in development)

How it Works

Letoplay works through a data collected by Dejavu audiofingerprinting. Check out dejavu documentation. The song is then broken off into time sections based on the bpm, which are used as targets for the data. To create data, a random sequence of fingerprints are chosen and given a label of the time section. The fingerprints are then called run through word2vec to create an numeric data. This data is then trained through a linear SVC kernel and tested with the test data to determine accuracy.

NOTE: The features described below are still under development. The specific training will be created based on tracking the fingerprints at specific times. The UI is developed through flask.

Using the Library

Installation

NOTE: pip install is in progress of being set up. For now just git clone this documentation. Follow installation instructions from Dejavu.

$ pip install -r requirements.txt

Running the Program

All Audio must be in .wav format. Place training files in the training folder and test files in the test folder. Delete any non-wav files from all train and test. The data folder must remain empty. Run model.py to check the accuracy score.

Using test.py, you can record your audio directly to a wav file which will be added to the test directory.

$ python3 test.py
File Name: firsttry
Seconds in Audio: 10
Recording...
Done Recording

To run the accuracy test:

$ python3 model.py

Expented Result:

              precision    recall  f1-score   support

     0.0_1.5       0.77      0.42      0.55        40
     1.5_3.0       0.80      1.00      0.89        40
   12.0_13.5       0.00      0.00      0.00         0
     3.0_4.5       0.84      0.93      0.88        40
     4.5_6.0       0.58      0.90      0.71        40
     6.0_7.5       0.91      1.00      0.95        40
     7.5_9.0       1.00      0.78      0.87        40
    9.0_10.5       1.00      0.65      0.79        40

    accuracy                           0.81       280
   macro avg       0.74      0.71      0.70       280
weighted avg       0.84      0.81      0.81       280

Accuracy Score:  0.8107142857142857

Development

Want to contribute or ask questions? Great! Email me at pradyunkumar03@gmail.com.