SPEECH

PROCESSING
Course Instructor:
Anjali Yadav
Assistant Professor
CSE Dept.
UNIT-III

Speech Compression:
Waveform Coders,
Linear Predictive Coders,
Analysis by Synthesis Technique,
Speech and Audio Coding Standards
Lossy Data Compression
Lossy data compression gives away the accuracy of some of its input data for a better
compression ratio. It is usually applied to multimedia files, where some loss of detail can be
tolerated. Some techniques include:
• Transform Coding: Uses mathematical transforms that shrink the data, usually in JPEG
• Quantization: Reducing the precision of data; it is common in image and video
compression.

Lossless Data Compression

Lossless data compression guarantees that the decompressed data is identical to the
original data. It works best for text and data files where precision matters.
• Huffman coding: Uses a frequency-sorted binary tree to locate values efficiently.
• Run-length encoding (RLE): This compresses sequences of replicated data values.
• Lempel-Ziv-Welch (LZW): It creates a dictionary of data patterns and replaces
them with shorter codes.
Run-Length Encoding (RLE)
Run-Length Encoding (RLE) is a simple and efficient method for compressing strings. It
works by replacing consecutive repeated characters with a count and the character itself. For
example, the string "AAAABBBCCDAA" can be compressed using RLE as
"4A3B2C1D2A".
• cA
• cA
Pulse Code Modulation
A technique by which analog signal gets converted into digital form in order to have signal
transmission through a digital network is known as Pulse Code Modulation. It is abbreviated
as PCM.

PCM systems are basically signal coders also known as waveform coders. PCM allows the
representation of the continuous time message signal as a sequence of binary coded pulses.
The binary form permits only 2 probable states i.e., 0 and 1.

The major steps involved in PCM is Sampling, Quantizing and Encoding.

Basics of PCM
In pulse code modulation, the analog message signal is first sampled, and then the amplitude
of the sample is approximated to the nearest set of quantization level. This allows the
representation of time and amplitude in a discrete manner. Thereby, generating a discrete
signal.
This discrete signal is then converted into its binary form for the transmission of the signal.

It is to be noted here that, in PCM technique the signal gets transmitted in the coded format
and must be decoded at the receiver in order to have the original message signal.
It is basically composed of a Transmitter, a Transmission Path and a Receiver.

The transmitter performs the sampling, quantizing and encoding of the signal.

The transmission path includes regenerative receivers that recover the signal
from the undesired noise effects.

Lastly, the receiver section that performs decoding of the coded signal after
regeneration of the signal at the receiver.
PCM Transmitter

LPF: Here, the message signal which is in the continuous time form, is allowed to pass
through a low pass filter (LPF).
This LPF whose cutoff frequency is fm eliminates the high-frequency components of the
signal and passes only the frequency components that lie below fm.

Sampler: The output of the LPF is then fed to a sampler where the analog input signal is
sampled at regular intervals. The sampling of the signal is done at the rate of f s. This
sampling frequency is so selected that it must follow the sampling theorem that is expressed
as:
Quantizer: A quantizer is a unit that rounds off each sample to the nearest discrete level.
The sampler provides a continuous range signal and hence still an analog one.
The quantizer performs the approximation of each sample thus assigning it a particular
discrete level.

As it basically rounds off the value to a certain level this shows some variation by the actual
amount. Thus we can say, quantizing a signal introduces some distortion or noise into it.
This is known as Quantization Error.

A quantizer can be of two types: uniform and non-uniform quantizer. In uniform quantizer,
there exists a uniform spacing in between the level. As against, in non-uniform quantizer, the
spacing in between the levels is not uniform.

Note: For a low signal level, the quantization error is high i.e., bad SNR. But, for a high
signal level, the quantization error is low providing good SNR.
Encoder: An encoder performs the conversion of the quantized signal into binary codes.
This unit generates a digitally encoded signal which is a sequence of binary pulses that acts
as the modulated output.

As it is a binary encoder thus generates a Binary Coded Sequence. That is transmitted

through the Transmission Path.
Transmission Path in a PCM system
PCM achieves low signal distortion by employing regenerative receivers along the
transmission path.
The channel introduces distortion in the signal during transmission. This distortion is
eliminated by the regenerator in order to provide a distortion less PCM signal. Resultantly,
enhancing the transmission ability of the system.

It basically performs equalization and timing and then executes decision making.
The PCM signal when provided to the regenerative repeater, the equalizer circuit at the
beginning performs the reshaping of the distorted signal. At the same time, the timing
circuit generates a pulse train that is a derivative of input PCM pulses.
This pulse train is then utilized by the decision-making device in order to sample the PCM
pulses. This sampling is done at the instant where maximum SNR can be achieved. In this
way, the decision-making device generates the distortion less PCM wave.

PCM Receiver

Regenerator: A regenerative repeater is placed at the receiving end also so as to have an

exact PCM transmitted signal. Here, also the regenerator works in a similar manner as that
when employed in the transmission path. It eliminates the channel induced noise and
reshapes the pulse.
DAC and Sampler: Digital to analog converter performs the conversion of digital signal again into its analog
form by making use of the sampler. As the actual message signal was analog thus at the receiver end there is a
necessity to again convert it into its original form.

LPF: The sampler generates analog signal but that is not the original message signal. Thus, the output of the
sampler is fed to the LPF having cutoff frequency fm. This is sometimes termed as the reconstruction filter that
produces the original message signal.

The process done at the transmitter is somewhat reversed at the receiver in order to generate the original analog
message signal.
Transmission bandwidth in Pulse Code Modulation
The transmission bandwidth of a PCM system is associated with a number of bits per sample.
If the number of bits per sample increases, the bandwidth also increases. In order to have a good
approximation, a large number of levels must be used but that will lead to a larger bandwidth requirement.

Consider each quantizer level is represented by ‘n’ binary digits. Then the levels represented by n binary digits
is given as:

Where: q is the digital level of the quantizer

Every sample is changed into n bits, thus, a number of bit per sample is ‘n’.
The number of samples per second is fs. Hence the number of bits per second which is also termed as
signalling rate is given as:

As transmission bandwidth is half the signalling rate, hence:

Therefore,

But we know, fs ≥ 2fm

Thus the bandwidth of the PCM system is given as:

Advantages of PCM
• Immune to channel induced noise and distortion.
• Repeaters can be employed along the transmitting channel.
• Encoders allow secured data transmission.
• It ensures uniform transmission quality.
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula
1. Articula