app.py

import streamlit as st
import numpy as np
import time
import pickle
import tensorflow as tf
from music21 import *
from keras.models import Sequential
from keras.layers import *
from midi2audio import FluidSynth
import shutil
import pretty_midi
import soundfile as sf  

def midi_to_audio(midi_file, output_file):
    # Load the MIDI file
    midi_data = pretty_midi.PrettyMIDI(midi_file)
    
    # Synthesize the audio from the MIDI data
    audio_data = midi_data.synthesize()
    
    # Save to a WAV file
    sf.write(output_file, audio_data, 44100)

####################### Music Generation Functions #######################
def generate(seq_len,x):
    """ Generate a piano midi file """
    #load the notes used to train the model
    with open('final_notes', 'rb') as filepath:
        notes = pickle.load(filepath)

    # Get all pitch names
    pitchnames = sorted(set(item for item in notes))
    n_vocab = len(set(notes))

    network_input, normalized_input = prepare_sequences(notes, pitchnames, n_vocab , seq_length = seq_len)
    model = create_network(normalized_input, n_vocab)
    prediction_output = generate_notes(model, network_input, pitchnames, n_vocab, x)
    create_midi(prediction_output)


def prepare_sequences(notes, pitchnames, n_vocab , seq_length):
    """ Prepare the sequences used by the Neural Network """
    # map between notes and integers and back
    note_to_int = dict((note, number) for number, note in enumerate(pitchnames))

    sequence_length = seq_length
    network_input = []
    normalized_input = []
    output = []
    for i in range(0, len(notes) - sequence_length, 1):
        sequence_in = notes[i:i + sequence_length]
        sequence_out = notes[i + sequence_length]
        network_input.append([note_to_int[char] for char in sequence_in])
        output.append(note_to_int[sequence_out])

    n_patterns = len(network_input)

    # reshape the input into a format compatible with LSTM layers
    normalized_input = np.reshape(network_input, (n_patterns, sequence_length, 1))
    # normalize input
    normalized_input = normalized_input / float(n_vocab)

    return (network_input, normalized_input)

def create_network(network_input, n_vocab):
    """ create the structure of the neural network """
    adam = tf.keras.optimizers.Adam(0.001)

    model = Sequential()
    model.add(LSTM(
        512,
        input_shape=(network_input.shape[1], network_input.shape[2]),
        recurrent_dropout=0.3,
        return_sequences=True
    ))
    model.add(LSTM(512, return_sequences=True, recurrent_dropout=0.3,))
    model.add(LSTM(256))
    model.add(BatchNormalization())
    model.add(Dropout(0.2))
    model.add(Dense(256))
    model.add(Activation('relu'))
    model.add(BatchNormalization())
    model.add(Dropout(0.2))
    model.add(Dense(n_vocab))
    model.add(Activation('softmax'))
     # 'rmsprop'
    model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])

    # Load the weights to each node
    model.load_weights('best2.h5')

    return model


def generate_notes(model, network_input, pitchnames, n_vocab , x):
    """ Generate notes from the neural network based on a sequence of notes """
    # pick a random sequence from the input as a starting point for the prediction
    start = np.random.randint(0, len(network_input)-1)

    int_to_note = dict((number, note) for number, note in enumerate(pitchnames))

    pattern = network_input[start]
    prediction_output = []

    # generate x notes (x entered by user)
    for note_index in range(x):
        prediction_input = np.reshape(pattern, (1, len(pattern), 1))
        prediction_input = prediction_input / float(n_vocab)

        prediction = model.predict(prediction_input, verbose=0)

        index = np.argmax(prediction)
        result = int_to_note[index]
        prediction_output.append(result)

        pattern.append(index)
        pattern = pattern[1:len(pattern)]

    return prediction_output


def create_midi(prediction_output):
    """ convert the output from the prediction to notes and create a midi file from the notes """
    offset = 0
    output_notes = []

    # create note and chord objects based on the values generated by the model
    for pattern in prediction_output:
        # pattern is a chord
        if ('.' in pattern) or pattern.isdigit():
            notes_in_chord = pattern.split('.')
            notes = []
            for current_note in notes_in_chord:
                new_note = note.Note(int(current_note))
                new_note.storedInstrument = instrument.Piano()
                notes.append(new_note)
            new_chord = chord.Chord(notes)
            new_chord.offset = offset
            output_notes.append(new_chord)
        elif pattern == 'r':
            # Create a rest note with a default duration (e.g., 1.0)
            new_note = note.Rest(1.0)  # Set a valid duration
            new_note.offset = offset
            new_note.storedInstrument = instrument.Piano()
            output_notes.append(new_note)
        else:
            new_note = note.Note(pattern)
            new_note.offset = offset
            new_note.storedInstrument = instrument.Piano()
            output_notes.append(new_note)
        offset += 0.5

    midi_stream = stream.Stream(output_notes)

    midi_stream.write('midi', fp='test_output2.mid')





    
# Set page config
st.set_page_config(page_title="Music Generation", page_icon=":tada:", layout="wide")


# Header section
with st.container():
    left_column, right_column = st.columns(2)
    with left_column:
        st.subheader("Music Generation :musical_keyboard:")
        st.write(
            "Our website is an application of piano music generation, you can listen to new musical notes generated by LSTM artificial neural network, which is used in fields of AI and deep learning. Let's get it started :notes:"
        )
    with right_column:
        # Display a GIF instead of Lottie animation
        st.image("im.gif", use_column_width=True)  

# Sidebar for user input
# Sidebar for user input
with st.sidebar:
    # Set a default value for len_notes
    default_len_notes = 100  # Example default value
    len_notes = st.slider('Please Choose The Notes Length', 20, 750, default_len_notes, 4)
    st.write("Notes Length = ", len_notes)

# Music generation functionality
if st.sidebar.button('Generate My Music'):
    # Use the default value if len_notes is not explicitly set by the user
    if len_notes is not None:
        with st.container():
            st.write("---")
            with st.spinner('✨ Your music is now under processing... ✨'):
                time.sleep(10)  # Simulate processing time
                generate(10, len_notes)
                midi_to_audio('test_output2.mid', 'output.wav')
                st.audio('output.wav')
                st.markdown("Here you are! You can download your music by right-clicking on the media player.")
    else:
        # Fallback to the default value if no selection is made
        with st.container():
            st.write("---")
            st.warning("No notes length selected. Using default value of 100.")
            with st.spinner('✨ Your music is now under processing... ✨'):
                time.sleep(10)  # Simulate processing time
                generate(10, default_len_notes)
                midi_to_audio('test_output2.mid', 'output.wav')
                st.audio('output.wav')
                st.markdown("Here you are! You can download your music by right-clicking on the media player.")