Audio%20Coding

1
Audio Coding

• Team Member
• ChungMing Yan, Chun Tong

2
Overview

• Mp3, AAC, Ogg Vorbis
• Technical specifications
• Test Results
• Sample clips
• Conclusion

3
Mp3

• MPEG1 Layer 3 Audio Coding
• A research project in EUREKA Digital Audio
  Broadcasting (DAB) in 1987
• A power of data reduction algorithm
• Standardized as ISO-MPEG Audio-Layer 3

4
Mp3 (Continued)

• Pros
• Fast Decoding
• Excellent hardware support
• ISO standard

• Cons
• Quality varies widely between encoders
• Even at highest quality, quality still suffers

5
Mp3 (Continue)

• Bit Rate Average 128 kbps or 192 kbps
• Sampling Frequency
• 16-24KHz (MPEG2 Layer 3)
• 32-48KHz (MPEG1 Layer 3)
• Parameters
• Birate
• 1. CBR (Constant Bitrate)
• 2. VBR (Variable Bitrate)
• 3. ABR (Average Bitrate)

6
(No Transcript)
7
Mp3 (Continue)

• Compression Techniques
• Huffman coding
• Non-linear quantization
• M/S Matrixing (Mid/side matrixing)
• Intensity stereo
• MDCT

8
Mp3 (Continue)

• Encoders Lame, Audio Catalysis
• Decoders Winamp, Window Media Player, etc.

9
AAC

• MPEG2/MPEG4 Advanced Audio Coding
• Developed by MPEG group (Dolby, Frauhofer, ATT,
  Sonyetc)
• More over of mp3

10
AAC (Continued)

• Pros
• Competitive at low and mid bitrates against other
  formats
• Decoders/Encoders work on all platforms

• Cons
• All high-quality implementations of AAC encoding
  are non-free and closed source.
• Relatively CPU intensive

11
AAC (Continue)

• Bit Rate 96 kbps, 128kbps, 196kbps
• Sampling Frequency48 full-bandwidth (up to 96
  KHz)
• Low Frequency Enhancement (LFE to 120 KHz)

12
(No Transcript)
13
AAC (Continue)

• Profiles
• LC (Low Complexity)
• Main
• Main LTP

14
AAC (Continue)

• Compression techniques
• Huffman coding
• Non-linear quantization and scaling
• Vector quantization
• M/S matrixing (middle/side channels) for high
  bitrates
• Intensity stereo for low bitrates
• TNS (temporal noise shaping)
• LTP(MPEG4 profile 2, reduce redundancy in
  successive frames)
• MDCT
• PNS (perceptual noise shaping)

15
AAC (Continue)

• Encoders Psytel AacEnc, Nero
• Decoders Winamp (with an AAC plug-in), QuickTime
  6

16
Ogg Vorbis

• Open source project
• Free, open, unpatented from other audio coding
  format

17
Ogg Vorbis (Continue)

• Pros
• Open source and patent free
• No loyalties even in commercial products

• Cons
• No commercial hardware players
• High bitrates not fully tuned

18
Ogg Vorbis (Continue)

• Bit rate 64kbps
• Sampling Frequency From 8 KHz (telephony) to
  192 KHz (Digital Masters)

19
Ogg Vorbis (Continued)

• Compression techniques
• Huffman coding
• MDCT (Cosine Sine)
• Wavelet in Vorbis II to improve quality

20
Ogg Vorbis (Continue)

• Encoders Besweet, OggDrop
• Decoders Winamp (with an Ogg Vorbis plug-in)

21
Test Result

• Three music clip used
• Orchestra
• Music with voice
• Voice only
• Different bitrate setting (switches)
• High bitrate
• Medium variable bitrate
• Low bitrate
• Additional switches (voice, pns)

22
Sample clips

• Wav
• Mp3
• Ogg
• AAC

23
filesizes

• mp3 track 1 track 2 track 3
• original 2470KB 2587KB 318KB
• cbr 256 (160 voice) 451 472 291
• r3mix (96) 283 381 174
• abr 48 83 91 90
• abr 32 54 60 60/60
• aac track 1 track 2 track 3
• original 2470KB 2587KB 318KB
• cbr 256 446 467 232
• abr 96 217 231 145
• abr 48 (tape, 40-59) 104 102 90
• cbr 32 57 60.4 59.2
• cbr 32 resampled 56.9 59.8 53.6
• ogg track 1 track 2 track 3
• original 2470KB 2587KB 318KB
• cbr 256 /8 389 427 304
• abr 96 /2 140 167 131

24
Switches used

• Mp3
• GUI automatically writes the proper command line
• CBR 256 - "c\EE3414\lame\lame.exe" -m s -b 256
  -k "C\EE3414\input.wav" "C\EE3414\output.mp3
• r3mix - "c\EE3414\lame\lame.exe" --nspsytune
  --vbr-mtrh -V1 -mj -h -b96 --lowpass 19.5
  --athtype 3 --ns-sfb21 2 -Z --scale 0.98 -X0
  "C\EE3414\input.wav" "C\EE3414\output.mp3"
• abr 48 - "c\EE3414\lame\lame.exe" --abr 48 -b 32
  -B 320/160 "C\EE3414\input.wav
  "C\EE3414\output.mp3
• abr 32 - "c\EE3414\lame\lame.exe" --abr 32 -b 32
  -B 320/160 "C\EE3414\input.wav"
  "C\EE3414\output.mp3"
• abr 32 voice - "c\EE3414\lame\lame.exe" --voice
  --abr 32 -b 32 -B 160 "C\EE3414\voice.wav"
  "C\EE3414\voice(abr32-voice).mp3"

25
Switches used (Continued)

• AAC
• cbr 256 -production -low_ath -profile 0 -br 256
• abr 96 -production -profile 0 -br 96 vbrhi
• abr 48 -tape
• abr 32 -br 32
• abr 32 -br 32 -resample 22050

26
Switches used (continued)

• Ogg Vorbis
• Lacking control parameters besides quality
• cbr 256 GUI quality set to 8
• abr 96 GUI quality set to 2
• abr 48 GUI quality set to -1
• abr 32 such bitrate is not possible with given
  tools even with manual bitrate tweaking

27
Conclusion

• ACC Ogg gt MP3
• There are very little differences, very hard to
  tell
• Depends on application
• Alternative Audio Coding
• Lossless encoding
• Monkey audio
• Speech specific
• Speez

28
Future research or improvements

• As technology improves, there will be newer
  coding schemes to be examined
• More extensive research of the parameters and
  encoding procedures
• Matlab waveform analysis (object analysis)
• Alternative Implementation

29
Resources

• http//www.audiocoding.com
• http//lame.sourceforge.net
• http//www.vorbis.com
• Team website
• http//chii.servehttp.com10240/ee3414