main_code.py

import numpy as np
import tensorflow as tf
from scipy.io.wavfile import write
import keras.backend as K
import librosa.display
import cv2
import librosa
import matplotlib.pyplot as plt
import librosa.display
import numpy as np
from keras.applications import VGG16
import os
import scipy

# Define function to preprocess input audio
#convert song to mel spectogram as siamese network doesn't work on sound directly
def create_spectrogram(clip,sample_rate,save_path):
    plt.interactive(False)
    fig=plt.figure(figsize=[0.72,0.72])
    S=librosa.feature.melspectrogram(y=clip,sr=sample_rate)
    librosa.display.specshow(librosa.power_to_db(S,ref=np.max))
    fig.savefig(save_path,dpi=400,bbox_inches='tight',pad_inches=0)
    plt.close()
    fig.clf()
    plt.close(fig)
    plt.close('all')
    del save_path,clip,sample_rate,fig,S
    
def load_img(path):
    img=cv2.imread(path)
    img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
    img=cv2.resize(img,(150,150))
    return img

import pickle

def main_loop():

    with open('dict.pickle', 'rb') as handle:
        songspecdict = pickle.load(handle)

    # Load the song to match
    song, sr = librosa.load("my_audio.wav")
    to_match = np.copy(song[0:220500])
    print("Loaded data into librosa...")

    # Create spectrogram image of the song to match
    create_spectrogram(to_match, sr, 'test.png')
    print("Created spectogram...")

    # Load the spectrogram image of the song to match
    to_match_img = load_img('test.png')
    to_match_img = np.expand_dims(to_match_img, axis=0)
    print("Loaded spectrum image...")

    # Get the embedding of the song to match
    # Load the tune recognition model
    model = tf.keras.models.load_model('./embdmodel_1.hdf5')
    embedding_model=model.layers[2]
    to_match_emb = embedding_model.predict(to_match_img)
    print("Get song embedding...")

    # Calculate the distances between the song to match and the songs in the database
    songsdistdict = {}
    for key, values in songspecdict.items():
        dist_array = []
        for embd in values:
            dist_array.append(np.linalg.norm(to_match_emb - embd))
            
        songsdistdict[key] = min(dist_array)
    song_titles=list(songsdistdict.keys())
    distances=list(songsdistdict.values())

    # Get the title and artist of the recognized song
    recognized_song_artist, recognized_song_title = song_titles[distances.index(min(distances))].split('-')
    recognized_song_title = os.path.splitext(recognized_song_title)[0]
    print(f'Artist: {recognized_song_artist}')
    print(f'Title: {recognized_song_title}')

    return recognized_song_title