Audio Signal Processing Perceptual Audio Coder

1
Audio Signal Processing-- Perceptual Audio Coder

• Shyh-Kang Jeng
• Department of Electrical Engineering/
• Graduate Institute of Communication Engineering

2
Encoder Basic Building Blocks
3
Decoder Basic Building Blocks
4
Time to Frequency Mapping Filter Banks

• Reduce redundancy
• Work with finite blocks of data to reduce delay
  and process memory requirements
• Use windows/overlapping to smooth out block
  discontinuities
• Prefer critically sampled transforms like the MDCT

5
Quantizer/Dequantizer

• Manipulate data using finite word length
  computers
• Quantization is the major source of distortion in
  digital audio coding (assuming adequate sample
  rate)
• Floating point quantizers can greatly increase
  average signal to noise for signals with wide
  dynamic ranges

6
Psychoacoustic Models

• Tell us that much of the signal is not
  perceptually relevant
• Masking models identify parts of the signal that
  can be ignored and regions where quantization
  noise will be inaudible
• Used to guide bit allocation
• Maximize the average difference between SNR and
  SMR

7
Bit Stream Formats

• Set by standards committees or publicized by
  vendors
• Describe the manner in which information is
  passed from the encoder to decoder
• The lower complexity one wishes the decoder to
  be, the more information must be passed through
  the bitstream

8
Standardizing Coder Technology

• Bit stream format and decoder mechanism are
  prescribed by a standard
• Encoder algorithms typically remain proprietary
• Standard bit stream formats allow audio files to
  be easily interchanged
• Allow multiple vendors to offer encoders that
  play on standard decoders

9
Sections in a Typical Bitstream Format

• Header information for bit alignment and to
  synchronizing timing
• Control parameters
• Bit allocation information
• Scale factors
• Data blocks

10
Header Information

• To identify the beginning of each block so the
  decoder knows where to begin extracting data
• Not needed at every block, just often enough so
  that the decoder does not lose track of the data
  locations
• Framing and synchronization data is used to
  synchronize the audio signal with accompanying
  video or other data
• Error checking codes (e.g., checksums, parity
  checks) may be included here.

11
Control Parameters

• To inform the decoder which of several coding
  options have been utilized
• Examples
• Block lengths
• Window types
• Type of algorithms used

12
Scale Factors

• Block floating point quantizing is used
• Often the blocks attempt to match the ears
  critical bands
• Scale factors for the data in each block need to
  be passed to the decoder
• Setting scale factors for blocks rather than for
  individual spectral samples greatly reduces the
  scale factor data

13
Bit Allocation Information

• Use psychoacoustic models
• Blockwise bit allocation is used to reduce the
  bits that must be passed to the decoder

14
Data Blocks

• Contain the mantissas of the spectral samples
• Usually interleaved by channel
• Cross-channel coding mechanisms sometimes use
  other bases for the multiple channels (e.g., sums
  and differences)