CANONICAL CORRELATION CNN FRAMEWORK FOR ROBUST NOISE SUPPRESSION FOR IMPROVED SPEECH QUALITY

Noise reduction is crucial for enhancing audio clarity in telecommunications and broadcast systems. Traditional methods like spectral subtraction have been widely used, leveraging noise spectrum estimation to suppress interference. While effective for stationary noise, these methods often introduce artifacts and struggle with dynamic noise conditions, limiting their applicability in real-time scenarios. This work proposes an Canonical Correlation Deep Learning (CCDL) for advanced noise reduction. The method dynamically adjusts its parameters to handle both stationary and non-stationary noise. Its Canonical Correlation Bandpass configuration focuses on the frequency range of the desired signal, ensuring superior noise suppression while preserving audio integrity. Real-time adaptability and artifact minimization further enhance its suitability for telecommunications and live broadcasting applications. Comprehensive evaluations demonstrate the proposed method's advantages over traditional techniques. Metrics such as Signal-to-Noise Ratio (SNR) and Perceptual Evaluation of Speech Quality (PESQ) validate its effectiveness in delivering clearer, more natural audio. The CCDL represents a significant advancement in noise reduction technology, addressing the limitations of existing methods and setting a new standard for audio processing in modern communication networks