model.py

# -*- coding: utf-8 -*-
"""Model.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1QBK60SxPNUE7AvxcXU7Essvmz0qBAcuP
"""

import pickle

# Load the pickle file
with open('hidden_states_by_genre.pkl', 'rb') as f:
    hidden_states_by_genre = pickle.load(f)

import pandas as pd
import numpy as np
from ast import literal_eval

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, accuracy_score

# Added for clustering
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler

embeddings = []
labels = []

for s in hidden_states_by_genre:
    for e in hidden_states_by_genre[s]:
        embeddings.append(np.array(e).flatten())
        labels.append(s)

# split data into train and test
X_train, X_test, y_train, y_test = train_test_split(
    embeddings, labels, test_size=0.2, random_state=42
)

# Standardize embeddings for both train and test
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)

# Perform clustering using KMeans on the training data
n_clusters = len(set(labels))  # Number of clusters equal to the number of genres
kmeans = KMeans(n_clusters=n_clusters, random_state=42)
train_clusters = kmeans.fit_predict(X_train_scaled)

# Apply the same KMeans transformation to the test data
test_clusters = kmeans.predict(X_test_scaled)

# Add clusters as additional features to both training and test sets
X_train_with_clusters = np.hstack([X_train_scaled, train_clusters.reshape(-1, 1)])
X_test_with_clusters = np.hstack([X_test_scaled, test_clusters.reshape(-1, 1)])

# Train random forest classifier
clf = RandomForestClassifier(n_estimators=100)
clf.fit(X_train_with_clusters, y_train)
preds = clf.predict(X_test_with_clusters)
probas = clf.predict_proba(X_test_with_clusters)

# Generate and print classification report
report = classification_report(y_test, preds)
print(report)

import librosa
import numpy as np
from sklearn.preprocessing import StandardScaler

# Load audio file and extract features
def extract_features_from_audio(file_path, target_dim=9984):
    y, sr = librosa.load("/content/The Police - Roxanne (Official Music Video).wav", duration=120)  # Load the audio (limit to 30 seconds)

    # Extract MFCCs (Mel-frequency cepstral coefficients)
    mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)  # You can adjust n_mfcc to control feature size

    # Flatten MFCCs
    mfccs_flattened = mfccs.flatten()

    # If the flattened features don't match the target dimension, resize or pad
    if len(mfccs_flattened) > target_dim:
        mfccs_flattened = mfccs_flattened[:target_dim]  # Truncate if larger
    elif len(mfccs_flattened) < target_dim:
        mfccs_flattened = np.pad(mfccs_flattened, (0, target_dim - len(mfccs_flattened)), 'constant')  # Pad if smaller

    return mfccs_flattened

# Example: Extract features from a music file and match target dimension
file_path = "/content/The Police - Roxanne (Official Music Video).wav"  # Replace with your audio file
new_embedding = extract_features_from_audio(file_path, target_dim=9984)

# Standardize the new embedding (using the same scaler used on the training data)
new_embedding_scaled = scaler.transform([new_embedding])

# Predict the cluster for the new embedding using the KMeans model
new_cluster = kmeans.predict(new_embedding_scaled)

# Add cluster as a feature
new_embedding_with_cluster = np.hstack([new_embedding_scaled, new_cluster.reshape(-1, 1)])

# Predict the genre using the trained classifier
new_pred = clf.predict(new_embedding_with_cluster)
print(f"Predicted genre: {new_pred[0]}")