Audio Data Preparation And Argument Using Tensorflow

# Audio Data Preparation, Augmentation & Classification Studio A comprehensive web application for analyzing, augmenting, and classifying audio data using **TensorFlow**, **YAMNet**, and **Flask**. ## 🚀 Features ### 1. Audio Ingestion - **File Upload**: Support for `.wav` file uploads. - **Microphone Recording**: Built-in browser recorder that captures audio, converts it to WAV client-side, and sends it for analysis. ### 2. AI Classification (YAMNet) - Integrates Google's **YAMNet** model. - Automatically identifies audio events from **521 classes** (e.g., "Speech", "Music", "Clapping", "Dog", "Siren"). - Displays top 5 predictions with confidence scores. ### 3. Advanced Visualization Pipeline The app processes audio through a multi-stage pipeline and visualizes each step: - **Waveform**: Raw time-domain signal. - **Trimming**: Automatic silence removal. - **Fading**: Logarithmic fade-in/out application. - **Spectrograms**: - Standard Spectrogram (Log Scale). - Mel Spectrogram (Perceptually relevant scale). - DB-Scaled Mel Spectrogram. - **Augmentations**: - **Frequency Masking**: Randomly masks frequency bands (useful for training robust models). - **Time Masking**: Randomly masks time steps. ## 🛠️ Tech Stack - **Backend**: Python, Flask - **ML/AI**: TensorFlow, TensorFlow I/O, TensorFlow Hub - **Audio Processing**: `tensorflow-io`, `soundfile`, `numpy` - **Frontend**: HTML5, CSS3 (Dark Mode), JavaScript (Web Audio API) ## 📋 Prerequisites - Python 3.9+ - Internet connection (for initial model download) ## ⚡ Installation 1. **Clone/Download** the project. 2. **Install Dependencies**: ```bash pip install -r requirements.txt ``` ## 🖥️ Usage 1. **Start the Server**: Run the following command in your terminal: ```bash python app.py ``` *Note: The first run may take a moment to download the YAMNet model.* 2. **Open Dashboard**: Go to [http://127.0.0.1:5000](http://127.0.0.1:5000) in your browser. 3. **Analyze Audio**: - **Upload**: Choose a `.wav` file and click **"Upload & Process"**. - **Record**: Click **"🔴 Record"**, speak or make a sound, then click **"⏹ Stop"**. ## 📂 Project Structure ``` ├── app.py # Main Flask application entry point ├── audio_processing.py # Core logic: TF pipeline, YAMNet classification ├── requirements.txt # Python dependencies ├── README.md # Documentation ├── uploads/ # Directory for temporary audio storage ├── static/ │ └── images/ # Generated visualization plots └── templates/ └── index.html # Dashboard frontend with Recorder logic ``` ## 🔧 Troubleshooting - **"Format not recognised" error**: Ensure you have refreshed the page to load the latest JavaScript recorder, which converts microphone input to WAV automatically. - **Model downloading**: If the app hangs on startup, check your internet connection; it needs to fetch the model from TF Hub. --- *Created with the assistance of Google DeepMind's Antigravity Agent.*