extract_features_from_songs.py

extract_features_from_songs.py

import librosa
import numpy as np
import pandas as pd
from tqdm import tqdm

def extract_features_from_a_song(x,sr):
    dict_features = {
        'zcr': np.mean(librosa.feature.zero_crossing_rate(x)),
        'chroma_stft': np.mean(librosa.feature.chroma_stft(x, sr=sr)),
        'mfcc': np.mean(librosa.feature.mfcc(x, sr=sr)),
        'spec_cent': np.mean(librosa.feature.spectral_centroid(x, sr=sr)),
        'spec_bw': np.mean(librosa.feature.spectral_bandwidth(x, sr=sr)),
        'spec_rolloff': np.mean(librosa.feature.spectral_centroid(x, sr=sr))
    }
    return dict_features

def get_data_from_sound(row):
    dict_features = {}
    dict_features['chroma_stft'] = 0
    dict_features['mfcc'] = 0
    dict_features['spec_cent'] = 0
    dict_features['spec_rolloff'] = 0
    dict_features['spec_bw'] = 0
    dict_features['zcr'] = 0
    audio_path = 'data/music/{}/{}.mp3'.format(row['style'],str(row['uuid']))
    exists = os.path.isfile(audio_path)
    if exists :
        x, sr = librosa.load(audio_path)
        uuid = row['uuid']
        try:
            pass
            #create_waveform_image(x, sr, uuid)
        except OverflowError:
            print('Cannot save waveform file for',uuid)
        try:
            pass
            #create_spectrogram_image(x,sr,uuid)
        except OverflowError:
            print('Cannot save spectogram file for',uuid)
        try:
            dict_features = extract_features_from_a_song(x,sr)
        except OverflowError:
            print('Cannot run features extract for',uuid)
        print('{} DONE'.format(uuid))
    row['spec_cent'], row['spec_rolloff'], \
    row['spec_bw'], row['mfcc'], row['zcr'], row['chroma_stft'] = dict_features['spec_cent'], \
                                                                  dict_features['spec_rolloff'], \
                                                                  dict_features['spec_bw'], \
                                                                  dict_features['mfcc'], \
                                                                  dict_features['zcr'], \
                                                                  dict_features['chroma_stft']
    
    return row
 
tqdm.pandas()
df = pd.read_csv(output_file,sep=";", header=0)
df = df.progress_apply(get_data_from_sound , axis=1)