Fusion Based Method for Wide Band Speech Reconstruction Over Legacy Telephone Networks

Abstract

Public Switched Telephone Networks (PSTNs) are limited to Restricted Band (RB) speech in the 0-4 kHz range, which reduces speech quality compared to Wideband (WB) speech (0-8 kHz). This study proposes a transform-based hybrid steganography technique combining Curvelet Transform (CT) and Fast Fourier Transform (FFT) to enhance RB speech quality while maintaining full PSTNs compatibility. The Curvelet Transform, capable of representing directional and edge-like features across multiple scales and angles, allows efficient capture of speech components such as formant transitions and unvoiced consonants, which are critical for improving quality and intelligibility. In the proposed approach, WB speech is decomposed into RB and Extended Band (4-8 kHz) components. Detailed coefficients of the RB are extracted using CT, and spread parameters of the Extended Band are embedded within the RB signal. Inverse CT and FFT are performed to form a Composite Restricted Band (CRB) signal for transmission. At the receiver, the embedded parameters are recovered to reconstruct the Extended Band, which is merged with the CRB signal to yield a high-quality Reconstructed Wideband (RWB) signal. Tests with participants aged 60-75 years demonstrated superior performance over conventional methods, with strong robustness to channel and quantization noise.