Speech Coding

Class Presentation of Custom DSP Implementation Course on:
Speech Coding
ECE Department - University of Tehran
Presented By:
Neda Kazemian Amiri
May 2005
This is a class presentation. All data are copy righted to
respective authors as listed in the references and have been used here for educational purpose only.

Outline
• Introduction
• Pulse Code Modulation (PCM)
• Algorithm Objectives and Requirements
• Vector Quantization vs. Scalar Quantization
• Linear Predictive Coding (LPC) and Speech Synthesis
• Speech Coding Schemes
• Measurement of Speech Quality
• ITU Encoding Standards

Introduction
The Spectrum of Human Speech [5]

Pulse Code Modulation (PCM)
The sampling and coding process and the resultant signal [5]

Algorithm Objectives and Requirements
• Quality and capacity
• Coding Delay
• Robustness
• Complexity and cost
• Voiceband Data Handling

Scalar Quantization
Uniform Quantization Non-Uniform Quantization [1]

Scalar Quantization
The technique of Predictive Quantization [5]

Vector Quantization
Partitioning of a 2-dimensional space to 18 cells
Full search Codebook [1]
X(n)
Input
Vector
Buffer
x
Vector
Matching
Codebook
Y
Index i
y

Vector Quantization
Codebook Types
• Full search codebooks
• Binary search codebooks
• Cascaded codebooks
• Split codebooks
• Gain shape codebooks
• Adaptive codebooks
• Random codebooks

Vector Quantization
Binary search codebooks
Binary Splitting into 8 cells [1]

Vector Quantization
Cascaded Codebooks
Cascaded Vector Quantizer [1]

Vector Quantization
Gain Shape Codebooks
Gain-Shape Vector Quantizer [1]

Voiced and Unvoiced Speech Waveforms [1]
Speech Signal for word “TO”

S(
z)
G
H ( z)
=
X ( z)
A(
z)
s(
n)
= Gx(
n)
+
p
= = − ∑
j=
z A(
) 1
1
Block Diagram of a Simplified Source Filter model of Speech Production [1]
p
∑
j=
1
a s(
n − j)
a
j
− j
j z

Linear Predictive Coding (LPC)
and Speech Synthesis [5]

Speech Coding Schemes [1],[2]

Measurement of Speech Quality
• Signal to Noise Ratio (SNR)
• Mean Opinion Scores (MOS)
MOS is a subjective measurement of speech quality coded in low bit rates.
1 = bad 2=poor 3 = fair 4 = good 5 = excellent

Quality Comparison of Speech Coding Schemes [2]

Block Diagram of a Sub-band Coder [1]
A Frequency Domain Hybrid Coder

Code Excited Linear Prediction
(CELP)
a) Coder and b) Decoder [3]

Adaptive Code Excited Linear
Prediction (ACELP) [3]
Adaptive Excitation Codebook

CELP Encoder with an Adaptive Codebook

Generalized Block Diagram of AbS LPC Coder
with Different Excitation types [1]

ITU Encoding Standards [7]
G.711: The first standard introduced for speech compression. It uses
PCM. Its bit rate is 64Kbps. It is used in PSTN.
G.721: It uses ADPCM. Its bit rate is 32Kbps. It is used in PSTN too.
G.722: It is like G.721. Its maximum bit rate is 64Kbps.
G.726: It uses ADPCM. Its bit rates are 16, 24, 32 and 40 Kbps.
G.723.1: It is a hybrid coder. With MP-MLQ algorithm its bit rate is
6.3 Kbps. With ACELP algorithm its bit rate is 5.3 Kbps. It is used for
videophones.
G.728: It is a hybrid coder with LD-CELP algorithm. Its bit rate is 16
Kbps. It uses 5 samples frames.
G.729: It is a hybrid coder with CS-ACELP algorithm. Its bit rate is 8
Kbps. It uses 10ms frames.

ITU Encoding Standards [7]

Comparison of Some Standards in the Terms
of Quality, Bit Rate and Frame Size

Some Characteristics of Speech
Compression Algorithms

Comparison of Some Standards in the Terms of Quality and Bit Rate

Summery and conclusion
• Some techniques of speech coding have been studied and
compared in terms of quality, bit rate, frame size, and
complexity.
• Algorithms with low bit rate, low complexity and fair Quality
are preferred.
• Quality will increase by increasing the bit rate in an
algorithm.
• Algorithms which are designed for low bit rate
communications, are more complex and thus consume more
power.
• Waveform coders such as G.711 and G.726 have low
complexity and introduce low delay.
• Hybrid coders have high complexity. G.728 is the most
complex algorithm, G.729 and G.723.1 have moderate
complexity.

References
[1] Kondoz, A. M., “ Digital Speech, Coding for Low Bit Rate
Communications Systems”, John Wiley & Sons, 1995.
[2] Xydeas, C., “An Overview of Speech Coding Techniques”,
Speech coding – Techniques and applications, IEE colloquium
on 14 Apr. 1992, pp. 111 – 125.
[3] Kipper, U., Reininger, H., and Wolf, D., “CELP Coding with
Adaptive Excitation Codebooks”, IEEE., 1991.
[4] Schroeder, M. R., and Atal, B. S., “Code-Excited Linear
Prediction (CELP): High-Quality Speech at Very Low Bit
Rates”, IEEE, 1985.
[5] Owen, F.E., “PCM and Digital Transmission Systems”,
McGRAW-HILL Book Company, 1976.
[6] Proakis, J. G., Salehi, M., “ Contemporary Communication
Systems Using MATLAB”, PWS Publishing Company, 1997
[7] Brunner, S., and Ali, A. A., “Voice over IP, Understanding
VoIP Networks”, Juniper Networks, Inc., 2004. www.juniper.net