Audio Bandwidth Extension Using Audio Super-Resolution

Abstract

Audio bandwidth extension (BWE) has emerged as an important tool for the satisfactory performance of low bitrate audio and speech codecs. In the existing BWE method, the high frequency (HF) excitation signals are generated by replicating the low frequency (LF) band directly. However, the coding perception quality will degrade if the correlation between LF and HF bands becoming weak. In this paper, we proposed a new algorithm to restore the HF excitation signals using audio super-resolution. The experiments shown the new algorithm have an outstanding performance for rebuilding HF excitation signals compare with the conventional replication method. In addition, we also provided a new BWE scheme based on audio super-resolution. According to our experimental results, in compare with LPC-based BWE, the subjective listening quality increased by 13% under the same bitrates; in compare with eSBR, the bitrates drop by 63.7% and have the approximate subjective listening quality.