Update readme.md

readme.md

@@ -1,4 +1,4 @@
-# 🌀 Washing-Machine Sound Classifier
+# Washing-Machine Sound Classifier
 
 This project implements a **two-stage hierarchical deep learning system** that classifies washing-machine sounds as **Normal** or **Abnormal**, and further identifies the **sub-class** (e.g., wash mode, spin mode, dehydration noise, bearing noise, etc.) using **Mel-spectrograms** and **CNN models**.
 
@@ -12,79 +12,71 @@ The project contains:
 
 ---
 
-## 📂 Project Structure
+## Project Structure
+
+├── app.py              # Gradio web app for prediction
+├── dl.py               # Training script (CNN models for Stage 1 + subclasses)
+├── extractaudio.py     # Script for testing audio files
+├── main_audio.py       # Dependencies list (duplicated from requirements)
+├── requirements.txt    # Project dependencies
+├── data_pipeline.py    # Data pipeline for spectrograms
+├── MelSpectrograms/    # Generated spectrogram dataset
+│   ├── 00-Abnormal/
+│   │   ├── 00-1 - Background noise/
+│   │   ├── 00-2 - Dehydration mode noise/
+│   │   └── 00-3 - Wash mode noise/
+│   └── 01-Normal/
+│       ├── 01-1 - Background noise/
+│       ├── 01-2 - Dehydration mode noise/
+│       └── 01-3 - Wash mode noise/
+└── saved_models/       # Saved trained models (.h5) + label_meta.json
 
-├── app.py # Gradio web app for prediction
-
-├── dl.py # Training script (CNN models for Stage 1 + subclasses)
-
-├── extractaudio.py # Script for testing audio files
-
-├── main_audio.py # Dependencies list (duplicated from requirements)
-
-├── requirements.txt # Project dependencies
-
-├── dl.py # Data pipeline for spectrograms
-
-├── MelSpectrograms/ # Generated spectrogram dataset
-
-│ ├── 00 - Abnormal/
-
-│ │ ├── Bearing noise/
-
-│ │ ├── Dehydration mode noise/
-
-│ └── 01 - Normal/
-
-│ ├── Wash mode/
-
-│ ├── Spin mode/
-
-└── saved_models/ # Saved trained models (.h5) + label_meta.json
 
 ---
 
-## ⚙️ Installation
+## Installation
 
 1) Clone the repository and move into it:
    ```bash
    git clone <repo-url>
    cd washing-machine-classifier
+   ```
 Create a virtual environment (recommended):
-
+```bash
 python -m venv .venv
 source .venv/bin/activate   # Linux/Mac
 .venv\Scripts\activate      # Windows
+```
 
 Install dependencies:
+```bash
 pip install -r requirements.txt
+```
 
 Data Preparation
 Place your raw audio dataset in the following structure:
 
-Washing machine/
-
-├── 00 - Abnormal/
-
-│   ├── Bearing noise/
-
-│   └── Dehydration mode noise/
-
-└── 01 - Normal/
-
-    ├── Wash mode/
-    
-    └── Spin mode/
-    
-Convert .wav files to Mel-Spectrograms:
-
+├── Washing machine/    # Generated spectrogram dataset
+│   ├── 00-Abnormal/
+│   │   ├── 00-1 - Background noise/
+│   │   ├── 00-2 - Dehydration mode noise/
+│   │   └── 00-3 - Wash mode noise/
+│   └── 01-Normal/
+│       ├── 01-1 - Background noise/
+│       ├── 01-2 - Dehydration mode noise/
+│       └── 01-3 - Wash mode noise/
+
+Convert .wav files to Mel-Spectrograms
+```bash 
 python dl.py
-This will generate the MelSpectrograms/ dataset.
+```
+This will generate the MelSpectrograms/dataset.
 
 Training Models
 Run:
-
+```bash 
 python dl.py
+```
 Trains Stage 1 model (Normal vs Abnormal).
 
 Trains Stage 2 models (sub-classes for Normal and Abnormal).
@@ -112,39 +104,42 @@ View prediction result and generated spectrogram in the browser.
 
 Example output:
 
-🎯 Final Prediction: Normal → Spin mode
-
-Confidence Scores:
---------------------
-Stage 1 (Normal): 0.9876
-Stage 2 (Spin mode): 0.9451
-Model Details
-Input: Mel-spectrogram images (224x224x3)
+🎯 Final Prediction: 00-Abnormal → 00-1 - Background noise 
 
-Backbone: Simple CNN (Conv2D + MaxPooling + Dense + Dropout)
+# Model Summary & Confidence Scores
 
-Stage 1: Normal vs Abnormal
+## Confidence Scores
+- **Stage 1 (Normal vs Abnormal):** 0.9876  
+- **Stage 2 (Sub-class within Normal, e.g., Spin mode):** 0.9451  
 
-Stage 2: Sub-classification within Normal/Abnormal
-
-Loss: Sparse categorical cross-entropy
-
-Optimizer: Adam
+---
 
-Metrics: Accuracy
+## Model Details
 
-Future Improvements
-Replace simple CNN with MobileNetV2 / EfficientNet for better accuracy.
+- **Input:** Mel-spectrogram images (`224×224×3`)  
+- **Backbone:** Simple CNN  
+  - `Conv2D + ReLU → MaxPooling`  
+  - `Flatten → Dense → Dropout → Dense → Softmax`  
+- **Stage 1:** Normal vs Abnormal classification  
+- **Stage 2:** Sub-classification within Normal / Abnormal  
 
-Add data augmentation (noise injection, pitch/time shift).
+### Training
+- **Loss:** Sparse Categorical Cross-Entropy  
+- **Optimizer:** Adam  
+- **Metrics:** Accuracy  
 
-Deploy as a FastAPI service for production.
+---
 
-Containerize with Docker.
+## Future Improvements
+- Replace simple CNN with **MobileNetV2** or **EfficientNet** for higher accuracy.  
+- Add **data augmentation**: noise injection, pitch/time shift, random time masking.  
+- Deploy as a **FastAPI** service for production.  
+- Containerize with **Docker** for easy deployment.  
 
-Requirements
-See requirements.txt:
+---
 
+## Requirements
+```text
 tensorflow
 librosa
 matplotlib
@@ -154,10 +149,6 @@ uvicorn[standard]
 python-multipart
 gradio
 
-👨‍💻 Author
-Developed by Anvit – Washing-machine sound anomaly detection with hierarchical deep learning.
-
----