音频编码中非均匀滤波器组的构造

为了更好地满足音频编码中时频映射的技术要求,基于频变调制重叠变换(FV-MLT),推导出更具一般性的频变扩展重叠变换(FV-ELT),从而构造了一类具有可变时频分辨率特性的滤波器组。它既可为瞬态音频信号提供灵活的时频映射,又保证稳态信号时有很高的频率分辨率。理论分析结果表明FV-ELT比MLT和FV-MLT有更高的变换增益,以8子带为例的仿真结果,也进一步验证FV-ELT提供了更好的时频分析特性。因此FV-ELT是非常适于音频编码时频分析的一类滤波器组,能够有效提高编码效率。

Nonuniform filter banks for audio encoding

Time-frequency mapping (TFM) of audio coding is improved by generalized FV-ELT (frequency-varying extended lapped transforms) derived based on FV-MLT (frequency-varying modulated lapped transforms). A class of nonuniform filter banks was constructed using FV-ELT with a variable time-frequency resolution. The filters provide more flexible TFM when applied to transient signals, while guaranteeing a higher frequency resolution for stationary signals. Theoretical analysis shows that FV-ELT give a higher transform gain than FV-MLT or MLT. Simulation results for an example with 8 nonuniform subbands shows that FV-ELT are superior to FV-MLT for some frequency properties. Thus, FV-ELT are a class of TFM filter banks for audio encoding with a higher encoding efficiency.