Audio Coding

  All lossy source coding techniques can be interpreted as vector quantization (VQ) with variable length coding . The coding scheme used in this coder uses an orthonormal transform (see 4.1.3) as a fixed codebook for vector quantization of MDCT (see 4.1.1) transform coefficients. Variable length coding is done by a number of Huffman codes on the VQ coefficients.

Since this is an experiment, the coder does not have a specified file format and does not output a bitstream. The produced bits are counted in the coder, and a new audio output file is produced with the resulting quantized data. The coder reads and writes raw wav files.

Before I move on to the actual coding, I introduce the transforms used in the coder.

  
Figure 6: A schematic view of the coder.

• Transforming
  • Modified Discrete Cosine Transform (MDCT)
  • A Fast MDCT Implementation
  • Karhunen Lòeve Transform (KLT)
• Quantization and Coding
  • Coefficient Clustering - Frequency Domain
  • Coefficient Clustering - Time Domain
  • Coefficient Quantization
  • Bit Coding
• Stereo Coding