Lossless Audio Compression

1
Lossless Audio Compression
2
Background

• Sound a wave
• Digital Audio samples of wave
• File structure header, L/R values
• CDs 44,100 samples/second
• Values 16 bit integers

3
Conversion To Digital Domain

• Original analog source truly lossless
• Conversion to digital (sampling) lossy
• No such thing as a digital microphone
• Sound quality based on sampling frequency and bit
  rate
• Nyquist Rate
• CD Red Book 44.1khz ,16 bit

4
Lossy Compression vs. Lossless Compression

• Lossy
• Approximation of original data
• Quality varies
• Original File 5009kb, Output File 120kb
• Sample
• Compression Ratio 2.39
• Compression Factor 41.74

5
Lossy Compression Techniques

• Perceptual Encoding
• Noise-Shaping
• Pre-Emphasis / De-Emphasis

Formats MP3, AAC, WMA,Ogg Vorbis, ATRAC
6
Lossy Compression Weaknesses

• Cannot recreate the original file exactly
• Successive generation degradation
• Lower sound quality

7
Lossy Compression vs. Lossless Compression

• Lossless
• Bit for bit reproduction of original data
• Original File 5009kb, Output File 3533kb
• Sample
• Compression Ratio 70.5
• Compression Factor 1.41

8
Why Lossless?

• Efficient storage
• Efficient transmission
• Maintain sound quality or Higher fidelity for
  given bit rate
• Archival level storage
• Absolute Transparency

9
Lossless Options

• Native format no compression
• Monkey
• FLAC
• WinZip
• Shorten

10
How Its Done

• Remove redundancy (compression)
• Channel Differences (L/R)
• Changes from One Sample to the next
• Shorten Bit Lengths

11
Results

• Short answer 60 to 70 of original file-size
  with pop, rock,  techno and other loud, noisy
  music 35 to 60 for quieter choral and
  orchestral pieces.
• http//www.firstpr.com.au/audiocomp/lossless/

12
(No Transcript)
13
(No Transcript)
14
http//www.monkeysaudio.com/
15
Step 1 Mid/Side Band Conversion

• Conversion to X,Y
• L/R Correlation
• Mid (X) (LR)/2
• Side (Y) (L-R)

Example L 12,500 R 12,800 --------------------
------------- Mid 12,650 Side 300
16
Background - Prediction Equations

• Predictor determination
• Previous values (x-1),(x-2), (x-3)

Shorten Prediction Equations P0 0 P1 X 1 P2
(2 X-1) X-2 P3 (3 X-1) (3 X-2)
X-3
17
(No Transcript)
18
Step 2 Predictor Determination

• Predicted values compared with the actual values
  and the difference (error) is what gets sent to
  the next state for coding

2,8,24,?PX (2 x-1) x-2 PX 2 24 8
40 ? 45 Passed Value 45-40 5
19
Step 2 Predictor Determination

• Adaptive Predictors
• m 0 to 1024 (none to full)
• After each prediction m is adjusted up or down
  depending on whether the prediction was helpful

2,8,24,? PX 40, ? 45, m 512 Final Value
PX
20
Step 3 Encoding of Data

• Goal make numbers as small as possible

Example 10,14,15,46 16 bit values 0000000000001
010 0000000000001110 0000000000001111 000000000010
1110 String value 101011101111101110
21
Step 3 Encoding of Data

• Statistical Compression Rice/Golomb Coding
• 1) k best guess at how many bits the number
  will take
• 2) x - the rightmost k bits of binary number
• 3) look at bits remaining after removing
  rightmost k bits, convert to decimal
• 4) use decimal number of zeros to encode number
  and terminate with a one
• 5) append result to x

• 10,14,15,4646 101110
• k (4 bits)
• k 1110
• 3) 10 binary 2 decimal
• 4) 001
• 5) 0011110

22
Decoding of Data

• Given 0011110
• 001 overflow 2 bits
• 2 (dec) 10 (bin)
• 10 1110 101110 46

23
Typical Lossless Compression for 650MB of Audio
24

• Bibliography
• http//www.firstpr.com.au/audiocomp/lossless/
• http//www.data-compression.com/
• http//www.monkeysaudio.com/
• Salomon, David. Data Compression The Complete
  Reference. New York Springer-Verlag, 2004.

25
Extra Information
26
Shorten Tony Robinson (1994)

• L/R channels separatedBlock size 256 samples
• Values converted to integers with a mean of 0
• Value predicted from previous values

27
Shorten Tony Robinson (1994)

• Insert scanned graphic

28
Shorten Tony Robinson (1994)

• Rice Coding

29
Shorten Tony Robinson (1994)

• P - predicted value
• A - actual value
• S - stored value
• S P A
• Block Size 128-256