main_json_create.py

import os
import librosa
import math
import json 
import matplotlib.pyplot as plt
import numpy as np

dataset_path = r"Data/genres_original" #location to gtzan data
json_path = r"data.json"
SAMPLE_RATE = 22050
DURATION = 30
SAMPLES_PER_TRACK = SAMPLE_RATE * DURATION

def save_mfcc(dataset_path, json_path, n_mfcc=13, n_fft=2048,
             hop_length=512, num_segments=5):
    # Data storage dictionary
    data = {
        "mapping": [],
        "mfcc": [],
        "labels": [],
    }
    samples_ps = int(SAMPLES_PER_TRACK/num_segments) # ps = per segment
    expected_vects_ps = math.ceil(samples_ps/hop_length)
    
    # loop through all the genres
    for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):
        # ensuring not at root
        if dirpath is not dataset_path:
            # save the semantic label
            dirpath_comp = dirpath.split("/")
            semantic_label = dirpath_comp[-1]
            data["mapping"].append(semantic_label)
            print(f"Processing: {semantic_label}")
            
            # process files for specific genre
            for f in filenames:
                if(f==str("jazz.00054.wav")):
                    # As librosa only read files <1Mb
                    continue
                else:
                    # load audio file
                    file_path = os.path.join(dirpath, f)
                    signal,sr = librosa.load(file_path,sr=SAMPLE_RATE)
                    for s in range(num_segments):
                        start_sample = samples_ps * s
                        finish_sample = start_sample + samples_ps
                        mfcc = librosa.feature.mfcc(y=signal[start_sample:finish_sample],
                            sr=sr,
                            n_fft=n_fft,
                            n_mfcc=n_mfcc,
                            hop_length=hop_length)

                        mfcc = mfcc.T

                        # store mfcc if it has expected length 
                        if len(mfcc)==expected_vects_ps:
                            data["mfcc"].append(mfcc.tolist())
                            data["labels"].append(i-1)
                            print(f"{file_path}, segment: {s+1}")

    with open(json_path,"w") as f:
        json.dump(data,f,indent=4)

save_mfcc(dataset_path, json_path, n_mfcc=13, n_fft=2048)