An application of spectral translation and spectral envelope extrapolation for high-frequency bandwidth extension of generic audio signals

The scope of this work is to introduce a conceptually simple yet effective algorithm for blind high-frequency bandwidth extension of audio signals, a means of improving perceptual quality for sound which has been previously low-pass filtered or downsampled (typically due to storage considerations). The algorithm combines an application of the modulation theorem for discrete Fourier transform to regenerate the missing high-frequency end of the signal spectrum with a linear-regression-driven approach to shape the spectral envelope for the regenerated band. The results are graphically and acoustically compared to those obtained with existing audio restoration software for a variety of input signals. The source code and Windows binaries of the resulting algorithm implementation are also included.