Upload 4 files

app.py

@@ -0,0 +1,122 @@
+### IMPORTS ###
+import os
+os.environ["KERAS_BACKEND"] ="tensorflow"
+import random
+import keras
+import gradio as gr
+from src.inference import generate_chorale, draw_random_sample
+from src.dataset import NoteEncoder
+from src.metrics import Preplexity
+from src.config import URL
+from src.utils import get_dataset_path, midi_to_wave, load_css, load_markdown
+
+
+### SETUP ###
+ROOT_DIR = os.getcwd()
+TRAIN_PATH, VAL_PATH, ARTIFACTS_PATH, MODEL_PATH = get_dataset_path(ROOT_DIR, URL)
+AUDIO_SAMPLES_PATH = os.path.join(ROOT_DIR, "samples")
+os.makedirs(AUDIO_SAMPLES_PATH, exist_ok=True)
+midi_path = os.path.join(AUDIO_SAMPLES_PATH, "sample.mid")
+wav_path = os.path.join(AUDIO_SAMPLES_PATH, "sample.wav")
+
+
+### DOWNLOAD SF2 MUSIC FONT ###
+sf2_download_path = keras.utils.get_file(
+        "FluidR3_GM.zip",
+        "https://keymusician01.s3.amazonaws.com/FluidR3_GM.zip",
+        extract= True,
+        cache_dir= ARTIFACTS_PATH,
+        cache_subdir= ""
+        )
+SF2_PATH = os.path.join(sf2_download_path, "FluidR3_GM.sf2")
+
+
+### LOAD MODEL & ENCODERS ###
+model = keras.models.load_model(os.path.join(MODEL_PATH, "bach_model.keras"),
+                                custom_objects={"Preplexity": Preplexity})
+note2id, id2note, vocab = NoteEncoder(vocab_path=ARTIFACTS_PATH, samples_path=None)
+
+
+### GRADIO ASSETS ###
+css = load_css()
+english_summary = load_markdown("english_summary")
+persian_summary = load_markdown("persian_summary")
+english_help = load_markdown("english_help")
+persian_help = load_markdown("persian_help")
+english_title = "# BachNet: AI-Generated Bach Music"
+persian_title = "# باخ‌نت: خلق موسیقی مشابه باخ با هوش مصنوعی"
+
+
+### GENERATION FUNCTIONS ###
+def pick_random_seed():
+    return draw_random_sample(VAL_PATH, seed=random.randint(0, 9999))
+
+def generate_fn(seed_path, seed_len, gen_len, temp):
+    sample_rows = slice(0, seed_len)
+    
+    generate_chorale(
+        model=model,
+        sample_seed_path=seed_path,
+        note2id=note2id,
+        id2note=id2note,
+        file_name=midi_path,
+        max_len=gen_len,
+        temperature=temp,
+        sample_seed_rows=sample_rows
+    )
+    
+    midi_to_wave(midi_file_path=midi_path, SF2_PATH=SF2_PATH, wave_path=wav_path)
+    return wav_path
+
+def set_english():
+    return (gr.update(value=english_title, elem_classes=[]),
+            gr.update(value=english_summary, elem_classes=[]),
+            gr.update(value=english_help, elem_classes=[]))
+    
+def set_persian():
+    return (gr.update(value=persian_title, elem_classes=['persian']),
+            gr.update(value=persian_summary, elem_classes=['persian']),
+            gr.update(value=persian_help, elem_classes=['persian']))
+
+
+### GRADIO APP ###
+with gr.Blocks(css=css, title="BachNet") as demo:
+    title_md = gr.Markdown(english_title, elem_id="title")
+    
+    with gr.Row():
+        english_btn = gr.Button("English")
+        persian_btn = gr.Button("Persian (فارسی)")
+    
+    summary_md = gr.Markdown(english_summary, elem_id="summary", max_height=None)
+    
+    with gr.Row(variant="panel"):
+        with gr.Column(scale=1, variant="panel"):
+            gr.Markdown("## Customize Your Chorale")
+            with gr.Row():
+                sample_seed_btn = gr.Button("Pick Random Seed", variant="primary")
+                seed_path_box = gr.Textbox(label="Selected Seed Path", interactive=False)
+            
+            seed_len_slider = gr.Slider(50, 150, 80, step=1, label="Seed Length")
+            gen_len_slider = gr.Slider(20, 200, 50, step=1, label="Generated Length")
+            temp_slider = gr.Slider(0.5, 1.8, 1.0, step=0.1, label="Temperature")
+            
+            generate_btn = gr.Button("Generate", variant="primary")
+        
+        with gr.Column(scale=1, variant="panel"):
+            gr.Markdown("## Generated Music: Listen & Download")
+            audio_player = gr.Audio(label="Generated Chorale", type="filepath",
+                                    interactive=False, show_download_button=True, streaming=True, autoplay=True)
+            help_md = gr.Markdown(english_help, elem_id="help_text")
+
+
+### EVENTS ###
+    sample_seed_btn.click(pick_random_seed, outputs=seed_path_box)
+    generate_btn.click(generate_fn, inputs=[seed_path_box, seed_len_slider, gen_len_slider, temp_slider],
+                       outputs=audio_player)
+    
+    english_btn.click(set_english, outputs=[title_md, summary_md, help_md])
+    persian_btn.click(set_persian, outputs=[title_md, summary_md, help_md])
+
+### LAUNCH APP ###
+if __name__ == "__main__":
+    demo.launch()

packages.txt

@@ -0,0 +1 @@
+fluidsynth

requirements.txt

@@ -0,0 +1,4 @@
+tensorflow==2.19.0
+numpy==2.1.3
+gradio==5.49.0
+music21==9.7.1

train.py

@@ -0,0 +1,33 @@
+from src.config import *
+from src.dataset import NoteEncoder, seq2seq_dataset
+from src.model import get_model
+from src.trainer import train_model
+from src.utils import get_dataset_path
+import keras
+import os
+
+### DOWNLOAD DATASET ###
+ROOT_DIR = os.getcwd()
+TRAIN_PATH, VAL_PATH, ARTIFACTS_PATH, MODEL_PATH = get_dataset_path(ROOT_DIR, URL)
+
+### REPRODUCABILITY ###
+keras.utils.set_random_seed(SEED)
+
+### INITIALIZE MODEL & DATASET ###
+note2id, id2note, vocab = NoteEncoder(samples_path=TRAIN_PATH, vocab_path=ARTIFACTS_PATH)
+vocab_size = len(vocab)
+
+train = seq2seq_dataset(TRAIN_PATH + "/*.csv",note2id, seq_len=SEQ_LEN, window_shift=WINDOW_SHIFT,
+                        batch_size=BATCH_SIZE, shuffle_buffer=2500, seed=SEED)
+
+val = seq2seq_dataset(VAL_PATH + "/*.csv" ,note2id, seq_len=SEQ_LEN, window_shift=WINDOW_SHIFT,
+                      batch_size=BATCH_SIZE, shuffle_buffer=None)
+
+bach_model = get_model(lr= LEARNING_RATE, weight_decay= WEIGHT_DECAY,
+                       emb_in = vocab_size, emb_out = EMBEDDING_DIM,
+                       lstm_layers = LSTM_LAYERS, lstm_units = LSTM_UNITS,
+                       lstm_dropout = LSTM_DROPOUT, dense_units = DENSE_UNITS,
+                       dropout = DROPOUT)
+
+### TRAINER ###
+train_model(bach_model, train, val, N_EPOCHS, ARTIFACTS_PATH, MODEL_PATH)