Spaces:
Sleeping
Sleeping
Rename and modify file
#1
by mandarmgd-03 - opened
- readme → readme.md +10 -32
readme → readme.md
RENAMED
|
@@ -29,34 +29,26 @@ The project contains:
|
|
| 29 |
│ ├── Spin mode/
|
| 30 |
└── saved_models/ # Saved trained models (.h5) + label_meta.json
|
| 31 |
|
| 32 |
-
yaml
|
| 33 |
-
Copy code
|
| 34 |
-
|
| 35 |
---
|
| 36 |
|
| 37 |
## ⚙️ Installation
|
| 38 |
|
| 39 |
-
1
|
| 40 |
```bash
|
| 41 |
git clone <repo-url>
|
| 42 |
cd washing-machine-classifier
|
| 43 |
Create a virtual environment (recommended):
|
| 44 |
|
| 45 |
-
bash
|
| 46 |
-
Copy code
|
| 47 |
python -m venv .venv
|
| 48 |
source .venv/bin/activate # Linux/Mac
|
| 49 |
.venv\Scripts\activate # Windows
|
| 50 |
-
Install dependencies:
|
| 51 |
|
| 52 |
-
|
| 53 |
-
Copy code
|
| 54 |
pip install -r requirements.txt
|
| 55 |
-
|
|
|
|
| 56 |
Place your raw audio dataset in the following structure:
|
| 57 |
|
| 58 |
-
css
|
| 59 |
-
Copy code
|
| 60 |
Washing machine/
|
| 61 |
├── 00 - Abnormal/
|
| 62 |
│ ├── Bearing noise/
|
|
@@ -66,16 +58,12 @@ Washing machine/
|
|
| 66 |
└── Spin mode/
|
| 67 |
Convert .wav files to Mel-Spectrograms:
|
| 68 |
|
| 69 |
-
bash
|
| 70 |
-
Copy code
|
| 71 |
python dl.py
|
| 72 |
This will generate the MelSpectrograms/ dataset.
|
| 73 |
|
| 74 |
-
|
| 75 |
Run:
|
| 76 |
|
| 77 |
-
bash
|
| 78 |
-
Copy code
|
| 79 |
python dl.py
|
| 80 |
Trains Stage 1 model (Normal vs Abnormal).
|
| 81 |
|
|
@@ -91,16 +79,12 @@ normal_model.h5
|
|
| 91 |
|
| 92 |
label_meta.json (class name order)
|
| 93 |
|
| 94 |
-
|
| 95 |
1) Test a single audio file
|
| 96 |
-
bash
|
| 97 |
-
Copy code
|
| 98 |
python extractaudio.py
|
| 99 |
Update the audio_file path in the script before running.
|
| 100 |
|
| 101 |
2) Run the Gradio Web App
|
| 102 |
-
bash
|
| 103 |
-
Copy code
|
| 104 |
python app.py
|
| 105 |
Upload a .wav file.
|
| 106 |
|
|
@@ -108,15 +92,13 @@ View prediction result and generated spectrogram in the browser.
|
|
| 108 |
|
| 109 |
Example output:
|
| 110 |
|
| 111 |
-
markdown
|
| 112 |
-
Copy code
|
| 113 |
🎯 Final Prediction: Normal → Spin mode
|
| 114 |
|
| 115 |
Confidence Scores:
|
| 116 |
--------------------
|
| 117 |
Stage 1 (Normal): 0.9876
|
| 118 |
Stage 2 (Spin mode): 0.9451
|
| 119 |
-
|
| 120 |
Input: Mel-spectrogram images (224x224x3)
|
| 121 |
|
| 122 |
Backbone: Simple CNN (Conv2D + MaxPooling + Dense + Dropout)
|
|
@@ -131,7 +113,7 @@ Optimizer: Adam
|
|
| 131 |
|
| 132 |
Metrics: Accuracy
|
| 133 |
|
| 134 |
-
|
| 135 |
Replace simple CNN with MobileNetV2 / EfficientNet for better accuracy.
|
| 136 |
|
| 137 |
Add data augmentation (noise injection, pitch/time shift).
|
|
@@ -140,11 +122,9 @@ Deploy as a FastAPI service for production.
|
|
| 140 |
|
| 141 |
Containerize with Docker.
|
| 142 |
|
| 143 |
-
|
| 144 |
See requirements.txt:
|
| 145 |
|
| 146 |
-
css
|
| 147 |
-
Copy code
|
| 148 |
tensorflow
|
| 149 |
librosa
|
| 150 |
matplotlib
|
|
@@ -153,12 +133,10 @@ fastapi
|
|
| 153 |
uvicorn[standard]
|
| 154 |
python-multipart
|
| 155 |
gradio
|
|
|
|
| 156 |
👨💻 Author
|
| 157 |
Developed by Anvit – Washing-machine sound anomaly detection with hierarchical deep learning.
|
| 158 |
|
| 159 |
-
yaml
|
| 160 |
-
Copy code
|
| 161 |
-
|
| 162 |
---
|
| 163 |
|
| 164 |
|
|
|
|
| 29 |
│ ├── Spin mode/
|
| 30 |
└── saved_models/ # Saved trained models (.h5) + label_meta.json
|
| 31 |
|
|
|
|
|
|
|
|
|
|
| 32 |
---
|
| 33 |
|
| 34 |
## ⚙️ Installation
|
| 35 |
|
| 36 |
+
1) Clone the repository and move into it:
|
| 37 |
```bash
|
| 38 |
git clone <repo-url>
|
| 39 |
cd washing-machine-classifier
|
| 40 |
Create a virtual environment (recommended):
|
| 41 |
|
|
|
|
|
|
|
| 42 |
python -m venv .venv
|
| 43 |
source .venv/bin/activate # Linux/Mac
|
| 44 |
.venv\Scripts\activate # Windows
|
|
|
|
| 45 |
|
| 46 |
+
Install dependencies:
|
|
|
|
| 47 |
pip install -r requirements.txt
|
| 48 |
+
|
| 49 |
+
Data Preparation
|
| 50 |
Place your raw audio dataset in the following structure:
|
| 51 |
|
|
|
|
|
|
|
| 52 |
Washing machine/
|
| 53 |
├── 00 - Abnormal/
|
| 54 |
│ ├── Bearing noise/
|
|
|
|
| 58 |
└── Spin mode/
|
| 59 |
Convert .wav files to Mel-Spectrograms:
|
| 60 |
|
|
|
|
|
|
|
| 61 |
python dl.py
|
| 62 |
This will generate the MelSpectrograms/ dataset.
|
| 63 |
|
| 64 |
+
Training Models
|
| 65 |
Run:
|
| 66 |
|
|
|
|
|
|
|
| 67 |
python dl.py
|
| 68 |
Trains Stage 1 model (Normal vs Abnormal).
|
| 69 |
|
|
|
|
| 79 |
|
| 80 |
label_meta.json (class name order)
|
| 81 |
|
| 82 |
+
Inference
|
| 83 |
1) Test a single audio file
|
|
|
|
|
|
|
| 84 |
python extractaudio.py
|
| 85 |
Update the audio_file path in the script before running.
|
| 86 |
|
| 87 |
2) Run the Gradio Web App
|
|
|
|
|
|
|
| 88 |
python app.py
|
| 89 |
Upload a .wav file.
|
| 90 |
|
|
|
|
| 92 |
|
| 93 |
Example output:
|
| 94 |
|
|
|
|
|
|
|
| 95 |
🎯 Final Prediction: Normal → Spin mode
|
| 96 |
|
| 97 |
Confidence Scores:
|
| 98 |
--------------------
|
| 99 |
Stage 1 (Normal): 0.9876
|
| 100 |
Stage 2 (Spin mode): 0.9451
|
| 101 |
+
Model Details
|
| 102 |
Input: Mel-spectrogram images (224x224x3)
|
| 103 |
|
| 104 |
Backbone: Simple CNN (Conv2D + MaxPooling + Dense + Dropout)
|
|
|
|
| 113 |
|
| 114 |
Metrics: Accuracy
|
| 115 |
|
| 116 |
+
Future Improvements
|
| 117 |
Replace simple CNN with MobileNetV2 / EfficientNet for better accuracy.
|
| 118 |
|
| 119 |
Add data augmentation (noise injection, pitch/time shift).
|
|
|
|
| 122 |
|
| 123 |
Containerize with Docker.
|
| 124 |
|
| 125 |
+
Requirements
|
| 126 |
See requirements.txt:
|
| 127 |
|
|
|
|
|
|
|
| 128 |
tensorflow
|
| 129 |
librosa
|
| 130 |
matplotlib
|
|
|
|
| 133 |
uvicorn[standard]
|
| 134 |
python-multipart
|
| 135 |
gradio
|
| 136 |
+
|
| 137 |
👨💻 Author
|
| 138 |
Developed by Anvit – Washing-machine sound anomaly detection with hierarchical deep learning.
|
| 139 |
|
|
|
|
|
|
|
|
|
|
| 140 |
---
|
| 141 |
|
| 142 |
|