diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..c9ebf2d
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,3 @@
+{
+    "python-envs.defaultEnvManager": "ms-python.python:system"
+}
\ No newline at end of file
diff --git a/quantum_acoustic_qsvm/README.md b/quantum_acoustic_qsvm/README.md
new file mode 100644
index 0000000..dfc6dbe
--- /dev/null
+++ b/quantum_acoustic_qsvm/README.md
@@ -0,0 +1,21 @@
+# Quantum Acoustic QSVM & Multi-Class QCNN Pipeline
+
+This repository contains a hybrid Quantum Machine Learning (QML) pipeline for acoustic signal processing, statistical feature filtering, and classification using PennyLane and Scikit-Learn.
+
+## 🔑 Key Features
+* **DSP & Spectral Extraction**: Fast Mel-Spectrogram extraction and feature aggregation using `librosa`.
+* **Statistical Feature Selection**: Vectorized $t$-test and ANOVA $F$-test filtering ($p < 0.01$) to compress frequency feature space.
+* **PennyLane QSVM Engine**: Vectorized state-vector embeddings (`lightning.qubit`) using an entangling feature map ansatz to precompute quantum Gram matrices for SVM classification.
+* **Hierarchical QCNN Architecture**: 8-qubit Quantum Convolutional and Pooling Neural Network built with custom variational ansatzes for 2D time-frequency patches and multi-class expectation decoding.
+
+---
+
+## 📁 Repository Structure
+
+```text
+quantum_acoustic_qsvm/
+├── README.md               # Project overview & usage instructions
+├── requirements.txt        # Dependencies
+├── main_pipeline.py        # Quantum Kernel + SVM pipeline script
+├── train_qcnn.py           # Multi-Class QCNN training script
+└── demo.ipynb              # Walkthrough demo notebook
\ No newline at end of file
diff --git a/quantum_acoustic_qsvm/demo.ipynb b/quantum_acoustic_qsvm/demo.ipynb
new file mode 100644
index 0000000..e69de29
diff --git a/quantum_acoustic_qsvm/main_pipeline.py b/quantum_acoustic_qsvm/main_pipeline.py
new file mode 100644
index 0000000..72e56b4
--- /dev/null
+++ b/quantum_acoustic_qsvm/main_pipeline.py
@@ -0,0 +1,94 @@
+import time
+import librosa
+import numpy as np
+import pennylane as qml
+from sklearn.model_selection import StratifiedKFold
+from sklearn.feature_selection import SelectKBest, f_classif
+from sklearn.preprocessing import MinMaxScaler
+from sklearn.svm import SVC
+from sklearn.metrics import accuracy_score
+
+# ==========================================
+# 1. GENERATE BASE DATASET (DSP FUNNEL)
+# ==========================================
+files = [librosa.example('nutcracker'), librosa.example('choice')]
+
+def extract_dsp_features(file_path):
+    y, sr = librosa.load(file_path, sr=None, mono=True)
+    mel_spec = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=32)
+    mel_db = librosa.power_to_db(mel_spec, ref=np.max)
+    return np.mean(mel_db, axis=1)
+
+base_0 = extract_dsp_features(files[0])
+base_1 = extract_dsp_features(files[1])
+
+def generate_dataset(size):
+    half = size // 2
+    X = np.vstack([
+        base_0 + np.random.normal(0, 0.5, size=(half, 32)),
+        base_1 + np.random.normal(0, 0.5, size=(half, 32))
+    ])
+    y = np.array([0] * half + [1] * half)
+    X_f = SelectKBest(score_func=f_classif, k=4).fit_transform(X, y)
+    X_q = MinMaxScaler(feature_range=(0, np.pi)).fit_transform(X_f)
+    return X_q, y
+
+# Standard operational dataset size
+X_data, y_data = generate_dataset(60)
+
+# ==========================================
+# 2. DEFINE QUANTUM KERNEL (WITH ENTANGLEMENT)
+# ==========================================
+n_qubits = 4
+dev = qml.device("lightning.qubit", wires=n_qubits)
+
+def ansatz(x, wires):
+    """Expressive quantum feature map with single-qubit rotations and entanglement."""
+    for i in range(len(wires)):
+        qml.RX(x[i], wires=wires[i])
+        qml.RY(x[i], wires=wires[i])
+    # Entangling layer
+    for i in range(len(wires)):
+        qml.CNOT(wires=[wires[i], wires[(i + 1) % len(wires)]])
+
+@qml.qnode(dev)
+def quantum_kernel_circuit(x1, x2):
+    ansatz(x1, wires=range(n_qubits))
+    qml.adjoint(ansatz)(x2, wires=range(n_qubits))
+    return qml.probs(wires=range(n_qubits))
+
+kernel_func = lambda x1, x2: quantum_kernel_circuit(x1, x2)[0]
+
+# ==========================================
+# 3. STRATIFIED CROSS-VALIDATION & HYPERPARAMETER TUNING
+# ==========================================
+print("📦 Computing Global Training Gram Matrix...")
+global_kernel_matrix = qml.kernels.square_kernel_matrix(X_data, kernel_func)
+
+param_grid = [0.1, 1.0, 10.0, 100.0]
+cv_strategy = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
+
+best_score = 0.0
+best_C = None
+
+print("⚡ Running Hyperparameter Optimization across Quantum Features...")
+for C in param_grid:
+    fold_scores = []
+    for train_idx, val_idx in cv_strategy.split(X_data, y_data):
+        # Slice Gram matrix correctly for precomputed SVM kernel
+        K_train = global_kernel_matrix[np.ix_(train_idx, train_idx)]
+        K_val = global_kernel_matrix[np.ix_(val_idx, train_idx)]
+        
+        clf = SVC(kernel="precomputed", C=C)
+        clf.fit(K_train, y_data[train_idx])
+        preds = clf.predict(K_val)
+        fold_scores.append(accuracy_score(y_data[val_idx], preds))
+        
+    mean_score = np.mean(fold_scores)
+    if mean_score > best_score:
+        best_score = mean_score
+        best_C = C
+
+print(f"\n🏆 Grid Search Optimization Complete!")
+print(f"Best Regularization Parameter ('C'): {best_C}")
+print(f"Mean Validation Cross-Validation Accuracy Score: {best_score * 100:.2f}%\n")
\ No newline at end of file
diff --git a/quantum_acoustic_qsvm/requirements.txt b/quantum_acoustic_qsvm/requirements.txt
new file mode 100644
index 0000000..d853811
--- /dev/null
+++ b/quantum_acoustic_qsvm/requirements.txt
@@ -0,0 +1,7 @@
+pennylane>=0.35.0
+pennylane-lightning>=0.35.0
+librosa>=0.10.1
+scikit-learn>=1.3.0
+scipy>=1.10.0
+numpy>=1.24.0
+matplotlib>=3.7.0
\ No newline at end of file
diff --git a/quantum_acoustic_qsvm/train_qcnn.py b/quantum_acoustic_qsvm/train_qcnn.py
new file mode 100644
index 0000000..994a668
--- /dev/null
+++ b/quantum_acoustic_qsvm/train_qcnn.py
@@ -0,0 +1,192 @@
+import time
+import librosa
+from pennylane import numpy as np
+import pennylane as qml
+from sklearn.preprocessing import MinMaxScaler
+from sklearn.metrics import accuracy_score, classification_report
+
+# ========================================================
+# 1. ENHANCED MULTI-CLASS 2D ACOUSTIC PATCH INGESTION
+# ========================================================
+files = [
+    librosa.example('nutcracker'), 
+    librosa.example('choice'),
+    librosa.example('trumpet'),
+    librosa.example('vibeace')
+]
+
+def extract_2d_spectrogram_patches(file_path):
+    """Extracts a static 4x4 frequency-time block preserving structural layout."""
+    y, sr = librosa.load(file_path, sr=None, mono=True)
+    mel_spec = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=4, n_fft=2048, hop_length=512)
+    mel_db = librosa.power_to_db(mel_spec, ref=np.max)
+    
+    mid_frame = mel_db.shape[1] // 2
+    patch_2d = mel_db[:, mid_frame:mid_frame + 4]
+    
+    if patch_2d.shape[1] < 4:
+        patch_2d = np.pad(patch_2d, ((0, 0), (0, 4 - patch_2d.shape[1])), mode='constant')
+        
+    return patch_2d 
+
+base_patches = [extract_2d_spectrogram_patches(f) for f in files]
+num_classes = 4
+
+def generate_multi_class_2d_dataset(size):
+    """Generates a dataset split evenly across 4 distinct audio classes."""
+    samples_per_class = size // num_classes
+    X_list, y_list = [], []
+    
+    for class_idx in range(num_classes):
+        for _ in range(samples_per_class):
+            noisy_patch = base_patches[class_idx] + np.random.normal(0, 0.2, size=(4, 4))
+            X_list.append(noisy_patch)
+            y_list.append(class_idx)
+            
+    X_raw = np.array(X_list)
+    y = np.array(y_list)
+    
+    X_flat = X_raw.reshape(size, 16)
+    X_scaled = MinMaxScaler(feature_range=(0, np.pi)).fit_transform(np.array(X_flat))
+    return X_scaled.reshape(size, 4, 4), y
+
+dataset_size = 40
+X_data, y_data = generate_multi_class_2d_dataset(dataset_size)
+
+target_map = {
+    0: np.array([0.8, 0.8]),
+    1: np.array([0.8, -0.8]),
+    2: np.array([-0.8, 0.8]),
+    3: np.array([-0.8, -0.8])
+}
+
+y_transformed = np.array([target_map[int(val)] for val in y_data])
+
+# ========================================================
+# 2. HIERARCHICAL QCNN WITH INDEPENDENT PAIR WEIGHTS
+# ========================================================
+n_qubits = 8 
+dev = qml.device("lightning.qubit", wires=n_qubits)
+
+def quantum_conv_layer_unshared(params, wires):
+    """Applies independent parameters to each qubit pair to break spatial symmetry."""
+    param_idx = 0
+    # Even pairs
+    for w1, w2 in zip(wires[0::2], wires[1::2]):
+        qml.CRX(params[param_idx], wires=[w1, w2])
+        qml.CRY(params[param_idx + 1], wires=[w2, w1])
+        param_idx += 2
+    # Odd pairs
+    for w1, w2 in zip(wires[1::2], wires[2::2]):
+        qml.CRX(params[param_idx], wires=[w1, w2])
+        param_idx += 1
+
+def quantum_pooling_layer(wires_sink, wires_source):
+    for si, so in zip(wires_sink, wires_source):
+        qml.CRZ(np.pi / 2, wires=[so, si])
+
+@qml.qnode(dev)
+def qcnn_circuit(patch_2d, weights):
+    features = patch_2d.flatten()
+    for i in range(n_qubits):
+        qml.RX(features[i], wires=i)
+        qml.RY(features[i + 8], wires=i)
+        
+    # Layer 1: 8 wires -> 11 parameters
+    quantum_conv_layer_unshared(weights[:11], wires=list(range(8)))
+    quantum_pooling_layer(wires_sink=[0, 2, 4, 6], wires_source=[1, 3, 5, 7])
+    
+    # Layer 2: 4 active wires -> 5 parameters
+    quantum_conv_layer_unshared(weights[11:16], wires=[0, 2, 4, 6])
+    quantum_pooling_layer(wires_sink=[0, 4], wires_source=[2, 6])
+    
+    return [qml.expval(qml.PauliZ(0)), qml.expval(qml.PauliZ(4))]
+
+# ========================================================
+# 3. OPTIMIZATION & EVALUATION
+# ========================================================
+def cost_function(weights, X_batch, y_batch):
+    loss = 0.0
+    for sample, target in zip(X_batch, y_batch):
+        expectations = qcnn_circuit(sample, weights)
+        loss += (expectations[0] - target[0]) ** 2 + (expectations[1] - target[1]) ** 2
+    return loss / len(X_batch)
+
+initial_lr = 0.15
+decay_rate = 0.85
+decay_patience = 3
+patience = 8
+
+np.random.seed(42)
+# Total of 16 independent parameters across conv blocks
+weights = np.random.randn(16, requires_grad=True)
+
+best_weights = weights.copy()
+best_loss = float('inf')
+patience_counter = 0
+
+print("🏋️ Starting Asymmetric QCNN Parametric Training Loop...")
+print("---------------------------------------------------------")
+
+epochs = 35
+batch_size = 8
+current_lr = initial_lr
+
+for epoch in range(epochs):
+    if epoch > 0 and epoch % decay_patience == 0:
+        current_lr *= decay_rate
+
+    opt = qml.AdamOptimizer(stepsize=current_lr)
+
+    indices = np.random.permutation(len(X_data))
+    X_shuffled = X_data[indices]
+    y_shuffled = y_transformed[indices]
+    
+    X_batch = X_shuffled[:batch_size]
+    y_batch = y_shuffled[:batch_size]
+    
+    weights, current_loss = opt.step_and_cost(lambda w: cost_function(w, X_batch, y_batch), weights)
+    
+    if current_loss < best_loss:
+        best_loss = current_loss
+        best_weights = weights.copy()
+        patience_counter = 0
+        status_msg = "⭐ (New Best Saved)"
+    else:
+        patience_counter += 1
+        status_msg = ""
+
+    print(f"Epoch {epoch+1:02d} | LR: {current_lr:.4f} | Loss: {current_loss:.4f} {status_msg}")
+
+    if patience_counter >= patience:
+        print(f"\n🛑 Early stopping triggered at Epoch {epoch+1}!")
+        break
+
+print("---------------------------------------------------------")
+print("🔮 Evaluation: Distance Decoding on Best Weights...")
+
+def decode_expectations_euclidean(exp_vector):
+    exp_arr = np.array(exp_vector)
+    best_class = 0
+    min_dist = float('inf')
+    
+    for class_idx, target_vec in target_map.items():
+        dist = np.sum((exp_arr - target_vec) ** 2)
+        if dist < min_dist:
+            min_dist = dist
+            best_class = class_idx
+            
+    return best_class
+
+final_predictions = []
+for sample in X_data:
+    raw_scores = qcnn_circuit(sample, best_weights)
+    predicted_class = decode_expectations_euclidean(raw_scores)
+    final_predictions.append(predicted_class)
+
+y_true = [int(val) for val in y_data]
+final_acc = accuracy_score(y_true, final_predictions)
+
+print(f"\n🏆 Post-Training Classification Accuracy: {final_acc * 100:.2f}%\n")
+print("📊 Detailed Classification Report:")
+print(classification_report(y_true, final_predictions, target_names=['Nutcracker', 'Choice', 'Trumpet', 'Vibe Ace'], zero_division=0))
\ No newline at end of file