Perceptual Audio Coding The AT

1
Perceptual Audio CodingThe ATT/Bell Labs view

• James D. Johnston
• Chief Scientist
• Neural Audio, Kirkland, Wa

2
The early work

• Harvey Fletcher, et al.
• Loudness curves
• Initial masking measurements
• Spatial hearing analysis
• Rabiner, Atal, Flanagan, Crochiere, Jayant et al.
• Digital Signal Processing advancements
• Lpc
• A(d)PCM
• Resampling
• Filtering
• CELP

3
My early work

• The commentary grade codec
• 56 kb/s, 7kHz bandwith, 16kHz sampling rate
  (pre-G-72x) 2-band SBC
• Good on most material
• Sounds awful on high passed material.
• Ok, what and why?
• Masking, or actually, a lack thereof.

4
PXFM

• A testbed for the Alliant FX8 computers
• Tonality metric in psychoacoustic model
• FFT overlap/add filterbank
• Bitstream compression
• First, used multiple-radix encoding
• Evolved into multiple Huffman codebooks for
  compression
• Simple M/S stereo (all or nothing)
• Ancestor of ASPEC
• And thus MP3

5
PAC

• Multichannel bitstream
• MDCT coder, 128/1024 size
• Pairwise channel coding for noise imaging control
• Huffman coding for bitstream,
• Sectioning
• Zero codebook
• Each channel pair has M/S coding on or off per
  scalefactor band
• The predecessor to AAC coding
• Won the bake off between BC and NBC codecs
• All features adapted into MPEG-2 AAC
• 1998 a2b Music from ATT Shut down for
  perceived lack of market.