Blind bandwidth extension of audio signals based on harmonic mapping in phase space

Abstract

Audio bandwidth extension can artificially restore the truncated high frequencies at the decoder from the transmitted low-frequency signal. This paper presents a new method for blind bandwidth extension of wideband audio based on harmonic mapping in phase space. First, the spectral series of audio signal is represented by using phase space reconstruction. Then, the harmonics of high frequencies are mapped out from the evolutionary trajectories of low-frequency components in phase space by using nonlinear prediction. Combining with spectral envelope estimation of high frequencies based on Gaussian mixture model, the bandwidth of wideband audio signals can be effectively extended to super wideband, without any side information. Subjective and objective testing results indicate that the proposed method improves the harmonic characteristics of the extended super-wideband audio and achieves a better quality than conventional blind bandwidth extension methods.