app.py

import gradio as gr
import librosa
import numpy as np
import tensorflow as tf
import pickle
from pydub import AudioSegment
import os


# === Load Model dan Label Encoder ===
model = tf.keras.models.load_model('/content/drive/MyDrive/AI CHORD RECOGNITION/Final/final_model.h5')
with open('/content/drive/MyDrive/AI CHORD RECOGNITION/Final/label_chord.pkl', 'rb') as f:
    label_encoder = pickle.load(f)

# === Konversi MP3 ke WAV ===
def convert_mp3_to_wav(mp3_path):
    sound = AudioSegment.from_mp3(mp3_path)
    wav_path = mp3_path.replace('.mp3', '.wav')
    sound.export(wav_path, format="wav")
    return wav_path

# === Konversi Audio ke Mel Spectrogram ===
def audio_to_mel_spectrogram(y, sr):
    y = librosa.util.normalize(y)
    mel_spectrogram = librosa.feature.melspectrogram(y=y, sr=sr, n_fft=2048, hop_length=512, n_mels=128)
    mel_spectrogram_db = librosa.power_to_db(mel_spectrogram, ref=np.max)
    mel_spectrogram_db = (mel_spectrogram_db + 80) / 80  # Normalisasi ke 0-1
    mel_spectrogram_db = tf.image.resize(mel_spectrogram_db[..., np.newaxis], (128, 128)).numpy()
    mel_spectrogram_db = np.repeat(mel_spectrogram_db, 3, axis=-1)
    return np.expand_dims(mel_spectrogram_db, axis=0)

# === Prediksi Chord ===
def predict_chords(audio_path):
    if audio_path.endswith('.mp3'):
        audio_path = convert_mp3_to_wav(audio_path)
    
    y, sr = librosa.load(audio_path, sr=22050)
    duration = librosa.get_duration(y=y, sr=sr)
    chords = []
    previous_chord = None

    for i in range(0, int(duration)):
        start_sample = i * sr
        end_sample = (i + 1) * sr
        y_segment = y[start_sample:end_sample]
        if len(y_segment) == 0:
            continue

        mel_spectrogram = audio_to_mel_spectrogram(y_segment, sr)
        prediction = model.predict(mel_spectrogram)
        predicted_index = np.argmax(prediction)
        predicted_chord = label_encoder.classes_[predicted_index]
        predicted_chord = predicted_chord.replace('_', '') 

        if predicted_chord != previous_chord:
            chords.append(predicted_chord)
            previous_chord = predicted_chord

    return f"Predicted Chords: {' - '.join(chords)}"

# === Gradio Interface ===
css = """
body {
    font-family: sans-serif;
}
.gradio-container {
    border-radius: 10px;
    box-shadow: 0 2px 5px rgba(0, 0, 0, 0.1);
    padding: 20px;
}
.gr-button {
    background-color: #4CAF50;
    color: white;
    border: none;
    padding: 6px 12px;
    font-size: 14px;
    cursor: pointer;
    border-radius: 5px;
}
.gr-button:hover {
    background-color: #4a894c;
}
.description {
    margin-top: 40px;
    margin-bottom:-10px;
}
"""

sample_audio_path = "/content/drive/MyDrive/AI CHORD RECOGNITION/Final/example.mp3"

title = f"AI Chord Recognition"
description = """
<div class='description'>
Upload an audio file (<strong>.mp3</strong> or <strong>.wav</strong>) and let the AI predict the chord progression. 
You can also try my sample audio.
</div>
"""

with gr.Blocks(css=css) as interface:
    gr.Markdown(f"<h1 style='text-align: center;'>{title}</h1>")  
    gr.Markdown("<h3 style='text-align: center;'>Hello! I'm Tigor Neilson Sinaga, with NIM 22.11.4725.<br>Welcome to my AI Chord Recognition project!</h3>")
    gr.Markdown(f"<h3 style='text-align: center;'>{description}</p>")
    
    audio_input = gr.Audio(type="filepath", label="Upload Audio (MP3/WAV)")
    use_sample_button = gr.Button("Use Sample Audio", size="sm")
    predict_button = gr.Button("Predict Chords") # Changed line
    output_text = gr.Textbox(label="Predicted Chords", lines=5, placeholder="Chord predictions will appear here...")
    
    use_sample_button.click(fn=lambda: sample_audio_path, inputs=[], outputs=audio_input)
    predict_button.click(fn=predict_chords, inputs=audio_input, outputs=output_text, queue=True)    
    gr.Markdown("<p style='text-align: center; font-size: 12px; color: grey;'>This project is still under development and has not yet reached high accuracy.</p>")

interface.launch(share=True)