NATIONAL POLYTECHNIC INSTITUTE

SCHOOL OF HIGHER MECHANICAL ENGINEERING AND

ELECTRIC UNIT ADOLFO LÓPEZ MATEOS
UNIT
ZACATENCO.

ENGINEERING IN COMMUNICATIONS AND ELECTRONICS

DIGITAL SIGNAL PROCESSING

DIGITAL PROCESS
OF VOICE AND IMAGE
Temporal Analysis of the voice

The characterization of the voice and the establishment of its temporal parameters
They are important, yes. Since they apply directly to the waveform of the signal, to
difference of the analysis methods in the frequency domain, which are based on
in the study of spectral representation. Some useful techniques and also
The simple ones for this analysis are the zero crossing rate, the energy, and the function.
of autocorrelation.

The voice signal shows large variations in amplitude (sounds

sonorous/sounds non-sonorous). The energy over short periods of time reflects well
these variations, allowing to distinguish between voiced and voiceless sounds. It is
evident that the key factor in temporal analysis is the Hamming window: its
shape and its duration. If the window is very long, then the energy will not change.
a lot from one segment to the next, so it will not reflect correctly the
energy changes of the signal. On the other hand, if it is very short, the energy will vary
too much, which won't help us either. What we are looking for is a step filtering.
low, in such a way that the energy has a relatively smooth signal, without
sharp variations, but that adequately represent the variation of the
energy of the voice.

Voice-noise discrimination

One of the main problems in voice analysis is being able to distinguish

when there is a voice signal and when there is none, as we mentioned earlier
non-sonorous sounds are similar to noise. The localization problem of
the beginning and end of a word is difficult, especially in recognition systems
of voice. The technique used is local energy and zero crossing rate.
When the beginning or end corresponds to a voiced sound, it is enough to calculate its
local energy, which will be far greater than that of the noise. If the word begins or
it ends with non-sonorous sounds, so the energy is very similar to that of
noise. However, non-sonorous sounds have a frequency content
much higher than that of
noise. Therefore, we will use the rate as a criterion to distinguish it from noise.
crosses by zero.

Sampling: The interest in voice coding or voice compression

it consists of obtaining a compact digital representation of the voice signal
to carry out an efficient transmission. Voice encoding
involves the sampling process. The sampling process is the process
that converts a continuous signal in time into a discrete signal,
furthermore, the infinite range of amplitudes is reduced to a finite set of
possibilities. The waveform of a sampled signal can be
represented in the following way:

Where isait is an analog signal, ñ is an integer and T is the period of

sampling or the time difference between two adjacent samples, which is

determined by the bandwidth or the highest frequency in a signal
as a starter.
Quantization binary representation converts an amplitude signal
continues in a discrete amplitude signal, with discrete values divided
in regions with uniform widths. To the difference between the non-entry
quantified and the quantified output is known as noise of
quantization. The distance between the amplitude levels is known as
step size and is generally represented by A. Each level of
Discrete amplitude n is represented by a code word c(n), for its
transmission. If we consider that all the discrete amplitude values of
a signal is represented by the same number of B bits and the frequency of
sampling is fs, we have that the bit rate transmitted in a channel
communication is given by:

Uniform quantization: Uniform quantizers are those that in the

the distance between all levels of reconstruction is equal,
Furthermore, no consideration is given to the signal that is going to
quantize.
Voice conversion to image

This technique aims to convert voice frames into images.

compressed, those that are subsequently transmitted, in order to then perform on the
receiver the synthesis process in order to recover the voice track. For
achieving it, the voice signal is processed through a bandpass filter bank,
which eliminate redundancy in the information. The filter bank is located
within the frequency range where the main formants are located
the voice those that are detected dynamically or are fixed arbitrarily
so that the frequencies at which these formants are located determine
the central frequencies of the bandpass filters. The signal filtered by the bank of
filters are subsequently processed using the FFT algorithm, such as the
Fourier transform delivers complex samples, the real and imaginary part of
the filtered signal in the frequency domain becomes each one by
separated in an image equivalent to those to which a format is applied
image compression such as the well-known JPEG, TIFF, and PNG, compressing
in this way the voice plot. Both images are the ones that are transmitted and are
received at the receiver, where the reverse process is carried out in such a way
obtain the signal in the time domain.
Voice transmission over data networks

In order to transmit the encoded voice samples over data networks, it is necessary to
it is necessary to assemble 'packages'. A voice channel consists of a flow of bits,

dependent on the codec used.

While a package could be formed with each voice sample, this would generate a
overhead too significant. On the other hand, if one waits for
"to gather" too many voice samples, to form a package with minimum
percentage overload, unacceptable delays can be introduced. A package
IP can have up to 1500 bytes of information. If with samples from the G.711 codec
If one wanted to complete the 1500 bytes of the IP packet, a delay would be introduced.
125µs x 1500
= 187.5 ms. This delay is not acceptable in conversation applications.
voice.

For this reason, 'windows' of 10 to 30 ms are generally taken. The samples

voice codings from each of these windows are "joined" and with them they
arman
packages. The size of these windows is configurable for some algorithms
of coding, and is standardized for others.

Bandwidth in IP for voice

Since sending voice over data networks requires assembling

"packages", the required bandwidth will depend on the "overhead" that
generate these packets. As has been seen, for the sending of voice over networks
packages use the RTP standard. This protocol is built on top of UDP,
which in turn is mounted on IP, which travels over Ethernet in the LAN. This
sum of protocols makes the bandwidth required for voice traffic
about Ethernet should be significantly greater than the audio bandwidth. Next
An example is presented for the G.711 codec.

For a 20 ms window, and using G.711 Law A audio encoding, one

they obtain 160 bytes of voice per frame

Voice frame bytes = 64 kb/s * 20 ms / 8 = 160 bytes

The IP packet (including the RTP and UDP protocols) adds 40 additional bytes.

IP packet bytes = 160 + 40 = 200 bytes

The Ethernet frame adds another 26 bytes:

Ethernet Frame Bytes = 200 + 26 = 226 bytes

Bartlane image transmission system
The Bartlane system is an image transmission technique invented in 1920.
to send encoded images through submarine cable lines between
London and New York.

Its name corresponds to its inventors, H. G. Bartholomew and M. D. McFarlane and

it was used for the first time to transmit an image across the Atlantic in
1921. The Bartlane system allowed for the transmission of a photograph between both points.

separated by 6000 km in just three hours.

In the transmitter, a perforated pattern was used on the telegraphic tapes.

through special devices and image decoding at the receiver
using telegraph printers equipped with suitable typefaces for the
system.
OSI MODEL
The OSI model establishes a hierarchical architecture structured in 7 layers. The
The idea is to break down the complex process of communication into several parts.

simpler problems and assign those problems to the different layers,

so that a layer does not have to worry about what the others do.
According to the hierarchical structure, each layer provides services to the layer

immediately superior, to the one that returns the obtained results, and in turn
it requests services from the immediately lower layer.

Below are the functions associated with each of the seven

levels of the OSI model.

Physical Level

It is the level that guarantees the transportation of information (in bits) through the medium.

transmission physical layer between two directly connected nodes.

The PDU at this level is the bit.

He is also responsible for the encoding/decoding of the data in

electrical signals and to establish the electric levels of the used signals,
transmission speed, size and shape of the connectors used, functions
and procedures for establishing and deactivating physical connections, etc.

Data Link Level

Provide the necessary functions to establish, maintain and release

reliable data link connections between network nodes with link of
connection
direct, ensuring flow control and error-free transmission of the blocks
of information.

To do this, it uses redundancy and control bits that will allow it to detect and
correct the errors introduced by the lack of reliability of the data circuits.

The PDU at this level is the Frame.

It also defines how transmission occurs in terms of size, structure,

content and sequence of the plots.

Network Level

It is the layer responsible for switching and routing functions, so

as of congestion control.

The PDU at this level is the Packet.

It is responsible for establishing, maintaining, and releasing network connections, providing

the precise procedures for data exchange between source and destination.
It will try to avoid packet loss in case of saturation or failure of certain
routes.

Data Entities in the OSI model

It is an element that processes information in an open system to apply it to

determined functions. In this case, it will deal with processing information for its
exchange between machines. A process is made up of:

Service Access Point (SAP): place where each layer interacts

the services of the layer that is just below

Interface Data Unit (IDU): information block that a layer

go to a lower layer

Protocol data unit (N-PDU): information packets that

transports the information that is intended to be sent over the network. This information

it will be fragmented and composed of a header that carries information about

 control. This information is exchanged between two entities that
they belong to the same level in different places.

Service Data Unit (SDU): each IDU consists of a field

of information for the control of the interface (ICI) and another field with

information with network information (SDU). An SDU of level n

it represents the PDU of level n+1, in this way n+1-PDU = n-SDU
 PROCESS OF
GENERATING
CODES OF
DATA
An automatic code generator is a tool that derives from
certain patterns, the source code of an application. The use of these
tools reduce the time needed for software development
it also ensures that the degree of programming errors remains
limited, thus reducing debugging and tuning times. The
4GL consists of procedures that generate source code based on what
expressed in the application design or data model. These tools
They are useful in areas where both pure code generation is needed, as well
also the creation of user interfaces or both. Within the scope of the
automatic code generation different approaches are raised about the
theme. There are proposals regarding the implementation of solutions based on
UML diagrams. With this approach, it aims to translate the schemes of
relationships between classes that represent a database to classes and interfaces
implemented in a particular language.

BASIC TERMINOLOGY IN DATA TRANSMISSION:

TERM UNITS DEFINITION

DATA ELEMENT BITS ONE OR ZERO BINARIES
DATA REASON BITS PER SECOND (bps) REASON WHY IT IS
THEY TRANSMIT THE
ELEMENTS OF
DATA
SIGNAL ELEMENT DIGITAL: A PULSE OF PART OF THE SIGNAL THAT
AMPLITUDE TENSION OCCUPY THE MOST INTERVAL
ANALOG CONSTANT: A SHORT
PULSE RATE, CORRESPONDING TO A
PHASE AND AMPLITUDE SIGNALING CODE
CONSTANTS
SIGNALING REASON NUMBER OF ELEMENTS REASON WHY IT IS
O SPEED OF PER SIGNAL PER SECOND TRANSMIT THE
MODULATION (BAUDIOS) ELEMENTS OF
SIGN

a digital signal is a sequence of discrete and discontinuous voltage pulses,

where each pulse is a signal element. Where binary data is
they transmit by encoding each bit of data in each signal element in the case
simpler
there will be a one-to-one correspondence between the bits and the signal elements. yes
all signal elements have the same algebraic sign (positive or
negative), the signal is said to be unipolar. in a polar signal, a logic state is
will be represented by a positive tension level and the other by a level
negative. The reason for the data of a signal is the transmission speed,
expressed in bits per second, at which the data is transmitted.

The duration or length of a bit is defined as the time taken in the

transmitter to emit a bit for a data rate r, the duration of a bit is 1/r,
while the modulation rate is the speed or rate at which it changes the
signal level which will depend on the coding scheme used tends to
to express in baud and is equivalent to one signal element per second.

DEFINITION OF DIGITAL SIGNAL CODING FORMATS:

non-return to zero (NRZ-L):

0 = nivel alto;
1 = low level.
inverted return to zero (nrzi):
0 = there is no transition at the beginning of the interval.

1 = transition at the beginning of the interval.

BIPOLAR-AMI:
0 = no signal.
1 = positive or negative level, alternating.
PSEUDOTERNARIA:
0 = positive or negative level, alternating.
1 = no signal