Voice Over Internet Protocol

1
Voice Over Internet Protocol

• Chapter 3
• Speech-Coding Techniques

2
Speech-Coding Techniques

• Voice Quality
• A little about speech
• Voice Sampling
• Quantization
• Types of Speech Coders
• G.711
• Adaptive Differential PCM (ADPCM)
• Analysis-by-Synthesis (Abs) Codecs
• G.728 Low-Delay CELP
• G.723.1 Algebraic Codec-Excited Linear Prediction
  (ACELP)
• G.729
• Selecting Codecs
• Cascading Codecs
• Tones, Signals, and Dual-Tone Multifrequency
  (DTMF) Digits

3
Voice Quality

• MOS (Mean Opinion Score)
• Excellence-5
• Good-4
• Fair-3
• Poor-2
• Bad-1
• Perceptual Speech Quality Measurement
• ITU P.861
• Evaluate speech coding

4
A little about Speech

• Vocal cord
• Vocal tract

5
??????

• ??????
• ?????????????,????????,???????????????????????????
  ??????
• ??,?????????????????????????,????????????????.
• ??????????,?????????,????????
• ????????????,????????????,???(??,bouncing),
  ??(????????????????????(????????)?
• ???????????????,????????????????

6
???(Digitization)

• ??????????????
• ????????????????????????,?????

7
???(Digitization)

• ?????????? ????????,???????????????????????
• ??????????????????,???????????????????
• ???????,?????????????????? ?????????????(sampling
  frequency) (???a)?
• ?????,????????8 kHz (8,000 samples per second)
  ?48 kHz. ???????????Nyquist??(?????)????
• ??????????????quantization(???b)?

8
(No Transcript)
9
???

• ???????????????,???????????
• ????????
• ???????????,??????????
• ?????????? (????)

10
Nyquist Theorem

• ???????????????,???????????????????????

11
Nyquist Theorem

• Nyquist?????????????????????????
• ??(a)??????????????????? (???????????????)?
• ???????????????,??(b)?????????????????????????
• ??,???????1.5?,??(c)???????????????? (alias)
  ,?????????????????? (????????????)?
• ?????????????????????,??????Nyquist rate?

12
(No Transcript)
13
(No Transcript)
14
Nyquist Theorem

• Nyquist Theorem ???????band-limited(????????????
  f1 ??????f2,??????????2(f2 - f1).
• Nyquist frequency Nyquist??????
• ???????????Nyquist frequency ???,??????????antiali
  asing filter?????????????????Nyquist
  frequency????
• ????,?????Alias Frequency?????
• falias fsampling - ftrue for ftrue lt fsampling
  lt 2 ftrue

15
??????Signal to Noise Ratio (SNR)

• ???????????????signal to noise ratio (SNR)
• ??????????
• SNR????decibel???(dB),?1 dB??1bel???????db???SNR??
  ?????????10?????,??????
• SNR 10log10V2signal /V2noise20log10Vsignal
  /Vnoise

16
??????Signal to Noise Ratio (SNR)

• ????????????????. ?? ????????????????,
  ?SNR??20log10(10)20dB.
• ?????,10????????????????????,???????SNR??10dB, or
  1B.
• ??????????????????decibel???,????????????????????,
  ??????????????

17
(No Transcript)
18
????????

• ????????????????????
• ???????(Non-uniform quantization)
  ?????????????????????????
• Webers Law???????????????????????
• ?Response8?Stimulus/Stimulus
• ???????k???,???????????????(???? r ????? s)
• dr k (1/s) ds with response r and stimulus s.

19
????????

• ????????????????r k ln s C, C ???????????
• ??????????r k ln(s/s0)
• s0 ??????????????(r 0 ?ss0).
• ?????????????????s?????????r??,???????????
• ??????????µ-law??,(?u-law)??????????????A-law,????
  ???????????
• ?????????????????

20
????????
21
µ100?255A 87.6???????????????
22
?????(filter)

• ????AD????,??????????????????????? ?????????????
• ?????,??????50Hz?10kHz??,??????????????????????ban
  d-pass filter?????????
• ????????????20Hz?20kHz??
• ?DA ??????,?????????
• ???????,?????????????????????step functions????
• ??????,?DA?????????low-pass filter?

23
?????????

• ???????????,??????????????????(Stereo)???????????
  , ??????

24
????????

• ??,???????????????? PCM (Pulse Code
  Modulation)???????? DPCM (??????????)??????????ADP
  CM?

25
Pulse Code Modulation

• ????????????????????????(sampling and
  quantization).
• ?????????????,??????????????????????????(?????)?

26
(No Transcript)
27
?????PCM

• ?????????????????????,?????????????
• ?????????????????????
• ??????????(extraneous).
• ???????????????low-pass filter,???????????????????
  ?????????

28
?????PCM

• ??????????????????????low-pass filtering,?????????
  ?,???(c)????

29
???????

• ????????PCM????,????????????????,???????????????,?
  ??????????????????
• ????????????????????? (temporal
  redundancy),???????,??????????????,???????0??????
  ??????(histogram)?

30
???????

• ????????,????????????????,???????????????
  ???????????????(spike)?
• ??????????????????,??????????????,???????????

31
?????????Lossless Predictive Coding

• Predictive coding ??????,?????????????????,??????
  ???,??????????
• ???????????,????PCM???????
• ????????????,??????????fn??? ???????????????? ?,
  ???????en???? ????????????

32
?????????Lossless Predictive Coding

• ?????????????????????fn-1, fn-2,
  fn-3?,??????????, ????????????

33
?????????Lossless Predictive Coding

• ???????????????????????
• ????(a)???8kHz????????,?????????8?????
• ???(b),????????????0????
• ???(c)????????????????????????0????
• ?????????????????0,??????????????????????????????

34
(No Transcript)
35
?????????

• ?? ??????????????0..255?,??????????-255..255?????
  ???????????????????????????????
• ????????????????????,??SU?SD, ????Shift-Up?Shift-
  Down????????????????
• ???????????????????????,??????????-15..16?????????
  ????????????????????(ShiftSU?SD)?,??????????-15..
  16?????
• ??100??????? SU, SU, SU, 4?

36
?????????

• Lossless predictive coding????????????????????????
  ???????????????

37
?????????

• ???????????,??????? f1, f2,f3, f4, f5 21, 22,
  27, 25, 22. ??, ????????????f1
  21,??????????????

38
DPCM

• Differential PCM???????????,??????????????
• ?????????????????????,???????????,?????Lloyd-Max
  ???, ?????????????????
• ???? ??? fn ??????, ?????, ?? ???????????

39
DPCM

• DPCM ????,???????????????? en,???????????
  ??DPCM???????
• ??????huffman coding??????????

40
DPCM

• ????????????????????????? ? ? ? ? ?
• ??????(distortion)???????? ? ? ? ? ? ?
  ?,????????????????????Lloyd- Max
  ??????????????????

41
DPCM

• ?????,???????????????????????????????,??????????,?
  ?????? ??????i??, ??????N?????,?????????????

42
Types of Speech Coders

• Waveform codecs
• High-quality output and not complex
• Large amounts of bandwidth
• Degrade significantly while using lower bandwidth
• Source codecs (vocoders)
• Match the incoming signal to a mathematical model
• Use linear predictive filter
• A set of model parameters replaces the signal
  itself
• Private communications such as military
  applications
• Hybrid codecs
• Perform a degree of waveform matching
• Utilize knowledge of how people produce sound in
  the first place

43
PCM

• Sample rate 8kHz
• Uniform quantization
• 12 bits per sample (96kbps)
• Non-uniform quantization
• A-law and µ-law
• 8 bits per sample (64kbps)
• MOS 4.3

44
ADPCM

• G.721 offer ADPCM-coded speech at 32Kbps
• G.721 has now superseded by G.726
• G.726
• A-law and µ-law
• Converted to 16Kbps, 24Kbps, 32Kbps(MOS 4.0) and
  40Kbps

45
Analysis-by-Synthesis (Abs) Codecs

• Hybrid coder can provide relatively acceptable
  quality at rate down to 16kbps
• Vocoder can provide intelligible speech at
  2.4kbps and lower
• Most successful and commonly used are time-domain
  Abs codec
• Linear prediction filter model for the vocal
  tract
• Linear Predictive Coding (LPC) vocoder
• Instead of using a simple two-state,
  voiced/unvoiced model
• Excitation signal is chosen by attempting to
  match to reconstructed speech waveform as closely
  as possible to the original speech
• MPE, RPE, CELP

46
G.728 Low-Delay CELP

• CELP(Code-Excited Linear Predictive)
• Filter
• Change over time
• A codebook of acoustic vector
• Each vector contain set of elements, which
  represent various characteristics of the
  excitation signals
• Transmit a set of information
• Filter coefficients, gain, and a pointer to the
  excitation vector chosen
• Sender and receiver have the same codebook
• G.728
• Operate on five sample at a time (sampled at
  8kHz)
• 1024 vectors, index to the vector is 10 bits
• 16kbps
• MOS 3.9

47
G.728 Low-Delay CELP encoder
48
G.728 Low-Delay CELP decoder
49
G.723.1

• Algebraic Code-Excited Linear Prediction (ACELP)
• 6.3 Kbps or 5.3 Kbps (8000Hz)
• Operate on 240 samples (30ms delay)
• 4 subframes of 60 samples
• Utilize a look-ahead of 7.5ms (totally 37.5ms)
• Silence suppression
• SID (Silence Insertion Description), 4 bytes
• MOS 3.8

50
G.729

• 8 kbps, a frame of 80 samples(8kHz)
• 5 ms look-ahead (totally 15ms)
• Frame size is 80bits
• MOS 4.0
• G.729 Annex A
• MOS 3.7
• G.729 Annex B
• G.729 Annex D
• G.729 Annex E

51
(No Transcript)
52
Selecting Codecs

• CDMA QCELP
• RFC 2658
• GSM (Enhanced Full Rate, EFR)
• RFC 1890
• Adaptive Multi-Rate (AMR)
• G.711 does not incorporate any logic to deal with
  loss
• G.729 have the capability to accommodate a lost
  frame by interpolating from previous frame

53
Cascaded Codecs

• Minimize the number of times that a given speech
  is coded and decoded
• In some cases, a VoIP implementation may be such
  that cascaded codecs are unavoidable
• Ensure that the quality does not degrade

54
Tones, Signals, and Dual-Tone Multifrequency
(DTMF) Digits

• Most sophisticated codec available today achieve
  bandwidth efficiency without losing significant
  quality due to smart algorithms and powerful DSP
• Based on how voice is produced in the first place
• Tones and beeps are needed to be transmitted
• Fax tones, various tones such as busy tones, and
  DTMF(two-stages dialing, voice mail retrieval,
  and other applications.
• G.711 can handle these tones, G.723.1, G.729
  cannot
• Use gateway to handle tones and speech in
  different ways
• IP network v.s. circuit switch network
• Speech (RTP) external signal(H.323, SIP)
• RTP data can represent the tone/event

55
RTP payload format for named events/tones