Perceptual Audio Coder

Assignment #4

Anserws to part b)

  • Data rates I = 256 kb/s/channel and 128 kb/s/channel.

    Sounds

  • Original signal (sum of 5 sinusoids) (.wav)
  • Quantized/Dequantized signal at 256 kb/s/channel (.wav) - rated 4.5 on ITU-R 5-point Impairment Scale
  • Quantized/Dequantized signal at 128 kb/s/channel (.wav) - rated 3.5 on ITU-R 5-point Impairment Scale
  • Quantized/Dequantized signal at 64 kb/s/channel (.wav) - rated 2.5 on ITU-R 5-point Impairment Scale

    Assignment #5

  • Bit allocation for the mantissas by hand.
  • Bit allocation for the mantissas using the masking model.

    Sounds

  • Original signal (sum of 5 sinusoids) (.wav)
  • Quantized/Dequantized signal at 256 kb/s/channel (.wav) - rated 4.2 on ITU-R 5-point Impairment Scale
  • Quantized/Dequantized signal at 128 kb/s/channel (.wav) - rated 4 on ITU-R 5-point Impairment Scale
  • Quantized/Dequantized signal at 96 kb/s/channel (.wav) - rated 3.8 on ITU-R 5-point Impairment Scale
  • Quantized/Dequantized signal at 64 kb/s/channel (.wav) - rated 2 on ITU-R 5-point Impairment Scale
    Note: It seems that the sounds done with uniform bit allocation sound better. It is probably because I use an algorithm that exploits ALL the bits in the case of the uniform allocation. The masking-based allocation loses a lot of bits and the coder could be improved to fix that problem.
    Last modified: Wed Feb 23 13:48:29 PST 2000