Design And Implement Of Low Bitrate Audio Bandwidth Extension

With the rapid development of Internet and Mobile Communication Technology, the amount of audio data is surging. The research of audio compression algorithm has been pay more attention by major organizations for standardization(such as MPEG、3GPP、ITU-T、AVS), the high-quality and low bit rate encoding technology becomes a hot spot. Audio bandwidth extension system as an important part of the audio coding technologies has also been the focus of scholars around the world.AVS P10 is China’s homegrown coding standard of mobile speech and audio, which provides a vivid coding sound quality at low bitrate. This thesis have analyzed the compression artifacts of the BWE technology that was used in AVS P10 standard, finding that, if the correlation between low band and high band is weak, the sub-band replication strategy used in it will lead to some coding artifacts, such as harmonic missing, harmonic redundancy and other issues. In order to suppress the compression artifacts, we proposed a new sub-band selection method——Pearson correlation coefficient method. Specifically: Calculating the correlation coefficients for each high-frequency and low-frequency sub-band one by one, finally choosing a maximum(the best correlation) sub-band and remember the index of sub-band, then the corresponding low band signals are copied to the high band. Repeat do this until all four high band signal is rebuild.The Pearson correlation coefficient method provides compression artifacts suppression, however, which brings an expensive calculation complexity. In order to reducing the coding complexity, we also provided the other BWE algorithm, which is a hybrid BWE scheme according to the different signal types. it is well known the spectrum envelope is an important parameter for subjective listening perception. For speech signal, the LPC coefficient is the best envelope representation, therefore, we select LPC-based BWE to coding speech-like signal; for music signal, the spectrum coefficient is the most envelope representation, therefore, we select FFT-based BWE to coding music-like signal.In this thesis, we have given the detail descriptions of Pearson correlation coefficient method and hybrid BWE scheme. The experiments illustrated the validity of our method compare with AVS P10. Compare with AVS P10, on the subjective listening testing, the Pearson correlation coefficient method and hybrid BWE increased 12.1% and 18% respectively, on the objective testing, increased 8% and 13.8% respectively. The provided two new BWE methods improved the coding sound quality of AVS P10, moreover, which also brings a new alternative BWE coding scheme.