Audio Signal Processing and Coding

1.2 A GENERAL PERCEPTUAL AUDIO CODING ARCHITECTURE

Over the last few years, researchers have proposed several efficient signal models (e.g., transform-based, subband-filter structures, wavelet-packet) and compression standards (Table 1.1) for high-quality digital audio reproduction. Most of these algorithms are based on the generic architecture shown in Figure 1.1.

The coders typically segment input signals into quasi-stationary frames ranging from 2 to 50 ms. Then, a time-frequency analysis section estimates the temporal and spectral components of each frame. The time-frequency mapping is usually matched to the analysis properties of the human auditory system. Either way, the ultimate objective is to extract from the input audio a set of time-frequency parameters that is amenable to quantization according to a perceptual distortion metric. Depending on the overall design objectives, the time-frequency analysis section usually contains one of the following:

• Unitary transform
• Time-invariant bank of critically sampled, uniform/nonuniform bandpass filters

  

  Figure 1.1. A generic perceptual audio encoder.

• Time-varying (signal-adaptive) bank of critically sampled, uniform/nonuniform bandpass filters
• Harmonic/sinusoidal analyzer
• Source-system analysis (LPC and multipulse excitation)
• Hybrid versions of the above.

The choice of time-frequency analysis methodology always involves a fundamental ...