telecommunications: speech and audio

15
CSN 08704 D ata,A udio,Video and Im ages http://asecuritysite.com/comm s Telecomm unications P rofBill Buchanan Audio and Speech

Upload: napier-university

Post on 13-Apr-2017

556 views

Category:

Education


3 download

TRANSCRIPT

Page 1: Telecommunications: Speech and Audio

CSN08704

Data, Audio, Video and Imageshttp://asecuritysite.com/comms

Telecommunications

Prof Bill Buchanan

Audio and Speech

Page 2: Telecommunications: Speech and Audio

Nyquist Sampling• Nyquist defined that we

can reconstruct a signal if we sample at twice the highest frequency.

• Speech: 4kHz – One sample every 125 μS.

• Audio: 20kHz - One sample every 25 μS.

Page 3: Telecommunications: Speech and Audio

Sampling and Quantisation• Sample at twice the

highest frequency of the signal.

• N bits gives 2N levels.

• Quality defined by SNR and Dynamic Range.

• Max error = +/- Full_scale/2N 000

001010011100101101110111

3 bits -> 8 levelsN bits -> 2N levels N2

scale Full21 =error Max

+ ADC 111 010 110 000

Clock (Twice highest frequency of signal)

Samples

Page 4: Telecommunications: Speech and Audio

Dynamic Range

minmaxrange Dynamic

VV

12levels ofNumber n

dB )12log(20 12

log20range Dynamicmax

max

nnV

V

if 2n is much greater that 1, then dB 02.6 2log20 2log20range Dynamic nnn

Page 5: Telecommunications: Speech and Audio

Dynamic Range

Number of bits DR (dB) [ratio] Number of bits DR (dB) [ratio] 1 6.02 [2] 11 66.23 [2 048] 2 12.04 [4] 12 72.25 [4 096] 3 18.06 [8] 13 78.27 [8 192] 4 24.08 [16] 14 84.29 [16 384] 5 30.10 [32] 15 90.31 [32 768] 6 36.12 [64] 16 96.33 [65 536] 7 42.14 [128] 17 102.35 [131 072] 8 48.16 [256] 18 108.37 [262 144] 9 54.19 [512] 19 114.39 [524 288] 10 60.21 [1 024] 20 120.41 [1 048 576]

Page 6: Telecommunications: Speech and Audio

Signal-to-Noise Ratio

dB 6.02+1.76=SNR nNumber of bits SNR (dB) [ratio] Number of bits SNR (dB) [ratio] 7 43.90 [156.68] 14 86.04 [20 044.72] 8 49.92 [313.33] 15 92.06 [40 086.67] 9 55.94 [626.61] 16 98.08 [80 167.81] 10 61.96 [1253.14] 17 104.10 [160324.5] 11 67.98 [2506.11] 18 110.12 [320626.9] 12 74.00 [5011.87] 19 116.14 [641209.6] 13 80.02 [10 023.05] 20 122.16 [1 282 331]

Link

Page 7: Telecommunications: Speech and Audio

Delta Modulation• 1 bit used to code.• Faster sampling rate.• Tracks signal.• Slope overload. This occurs when the

signal changes too fast for the modulator to keep up. It is possible to overcome this problem by increasing the clock frequency or increasing the step size.

• Granular noise. This occurs when the signal changes slowly in amplitude. The reconstructed signal contains a noise which is not present at the input.

DAC

+

-

Clock

Up/Down

InputOutputSample

and hold

Up/downcounter

1111111000100011000010101

AnalogueSignal

Decoded output

Code:1111111000100011000010101

Analoguesignal

DAC output

PCM

Slope overload

Input signalPCM

Reconstructed signal

Page 8: Telecommunications: Speech and Audio

ADM and DPCM• Adaptive Delta Modulation.

Change bit change to keep up with slope.

• Differential PCM. Quantise within the maximum change in level.

Analoguesignal

m levels

n levels

coding region

Currentsample

Nextsample

Input

n-bit bus

DifferentialPCM

DifferentialPCM

Analogueoutput

Low-passfilter

+

-

DAC Clockdelay

ADC

Low-passfilter

Sample andhold

+

-DAC

Page 9: Telecommunications: Speech and Audio

CSN08704

Data, Audio, Video and Imageshttp://asecuritysite.com/comms

Telecommunications

Prof Bill Buchanan

Speech Encoding

Page 10: Telecommunications: Speech and Audio

Speech Encoding• Subjective and system

tests have found that 12-bit coding is required to code speech signals, which gives 4096 quantization levels.

• Noise in speech more noticeable on low volumes.

Softspeech

Loudspeech

Quantization noise

Quantization noise

Quantization noise less noticeable because signalstrength swamps the quantizationnoise

Quantization noise noticeable

Page 11: Telecommunications: Speech and Audio

A-Law and μ-Law Encoding• Compander used to convert

12-bit samples into 8 bits.• Expander used to convert 8

bits into 12-bits.

000000000000

111111111111

11111111

00000000

Input code

Output code

Low-passfilter Sampler 12-bit

ADC Compander

8 kHz

Low-passfilter

12-bitDAC Expander

64 kbps

Input

Output

12-bitsamples

Output

Input

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

A=1

A=100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Output

Input

=1

=50=255

0for )1log()1log(

xxxy

11for

10for

log1)log(1

log1x

A

Ax

AAxA

Ax

y

Page 12: Telecommunications: Speech and Audio

Piecewise Linear Companding

1615.5

3247.5

6411.5

2049 A-Law4079.5 -Law

Input

Output

16

32

48

128

Segment 0

Segment 1

Segment 2

Segment 7

Input Companded Decoder level Decoded level number

Step size

0–1 … 15–16

000 0000 … 000 1111

0 … 15

0.5 … 15.5

1

16–17 … 31–32

001 0000 … 001 1111

16 … 31

16.5 … 31.5

1

32–34 … 62–64

010 0000 … 010 1111

32 … 47

33 … 63

2

64–68 … 124–128

011 0000 … 011 1111

48 … 63

66 … 126

4

128–136 … 248–256

100 0000 … 100 1111

64 … 79

132 … 252

8

256–272 … 496–512

101 0000 … 101 1111

80 … 95

264 … 504

16

512–544 … 992–1024

110 0000 … 110 1111

96 … 111

528 … 1008

32

1024–1088 … 1984–2048

111 0000 … 111 1111

112 … 127

1056 … 2016

64

Page 13: Telecommunications: Speech and Audio

Audio Encoding StandardsITU standard Technology Bit rate Description G.711

PCM 64 kbps Standard PCM

G.721 ADPCM 32 kbps Adaptive delta PCM where each value is coded with 4 bits

G.722 SB-ADPCM 48, 56 and 64 kbps Subband ADPCM allows for higher-quality audio signals with a sampling rate of 16 kHz

G.728 LD-CELP 16 kbps Low-delay code excited linear prediction for low bit rates

+ ADC Rate = 8 bits x 8 kHz= 64 kbps

8kHz

Samples

Page 14: Telecommunications: Speech and Audio

Time Division MultiplexingBits per time slot = 8 Number of time slots = 32 Time for frame = 125s

kbps 204810125832

Timebits of NorateBit 6

30

0 1 2 3 14 15

0 1 2 3 16 31

Speech 0 Speech 30

One multiframe every 2 ms

Time slot 0 - Frame word alignmentTime slot 16 - Signalling information

125 s

Page 15: Telecommunications: Speech and Audio

CSN08704

Data, Audio, Video and Imageshttp://asecuritysite.com/comms

Telecommunications

Prof Bill Buchanan

Audio and Speech