🎧 Audio Denoising and Enhancement with Spectral Gating (PyTorch + Torchaudio)

This script removes background noise and enhances the quality of a .wav audio file using spectral gating in PyTorch and Torchaudio.

---

🚀 Features

• 📥 Takes a noisy .wav file as input
• 🧠 Applies spectral gating to suppress background noise
• 🎧 Outputs a cleaned and enhanced .wav file
• ⚡ Runs on GPU (CUDA) if available

---

🛠️ How It Works

1. Load Audio: Uses torchaudio.load() and resamples to 16kHz (default for models like Whisper).

2. Spectral Gating: Converts waveform to spectrogram → estimates background noise → subtracts it → reconstructs waveform.

3. Save Clean Audio: Stores the denoised waveform as a .wav file.

---

🧱 File Structure

audio_denoise_enhance.py   # Main script
input.wav                  # Example input (your noisy audio)
output.wav                 # Output (cleaned audio)

---

▶️ How to Run

1. Install dependencies:

pip install torch torchaudio

Make sure your system has a GPU with CUDA if you want GPU acceleration.

---

2. Run the script:

python denoise.py input.wav output.wav

• input.wav: Path to your noisy audio file
• output.wav: Path where cleaned audio will be saved

---

📌 Notes

• The first 0.5 seconds of the input audio are assumed to contain background noise only (used for noise estimation).
• Spectral gating helps suppress constant low-level noise like fan hum, air conditioning, etc.
• You can modify n_fft or hop_length in the spectral_gate_denoise() function for finer control.

---

🧪 Example

python denoise.py noisy_speech.wav cleaned_speech.wav

---

🧠 What is Spectral Gating?

Spectral gating is a noise reduction technique that suppresses low-energy (noisy) frequencies in a spectrogram while preserving high-energy components (speech/music).

---