Speech Processing

This repository contains code for various speech processing tasks.

Subfolders:

• speech_preprocessing: Contains functions for preprocessing speech audio, including:
  • Spectrogram calculation
  • Noise cancellation (using spectral subtraction)
• wavs: Stores audio files used in the examples.

Files:

• Noise_cancelling.py: Implements functions for calculating spectrograms, noise cancellation, and audio reconstruction.
• speech_segmentation.py: Implements a basic speech segmentation engine using energy-based detection.

Noise Cancellation

The Noise_cancelling.py file provides functions for:

• Spectrogram calculation (spectrogram and spectrogram2wav) for converting between time and frequency domains.
• Noise cancellation (NoiseCancelling) using spectral subtraction to remove noise from an audio signal.

Speech Segmentation

The speech_segmentation.py file demonstrates a simple speech segmentation engine based on energy levels. The Engine class implements methods for:

• Calculating energy of the audio signal.
• Updating a dynamic threshold based on background noise.
• Detecting speech segments based on energy exceeding the threshold for a minimum duration.

Running the Examples

This repository requires libraries like numpy, matplotlib.pyplot, scipy.io.wavfile, and IPython.display (for audio playback). Make sure you have them installed before running the scripts.

The provided code snippets showcase the functionalities. You can modify them to suit your specific needs.

Future Work

This repository serves as a starting point for speech processing tasks. Here are some potential areas for future development:

• Implementing more sophisticated noise cancellation techniques.
• Utilizing feature extraction methods for speaker recognition or speech classification.
• Refining the speech segmentation engine for better accuracy.

Feel free to explore, modify, and contribute to this repository!