LPC Models and Different Speech Enhancement Techniques- A Review

NOVATEUR PUBLICATIONS
INTERNATIONAL JOURNAL OF INNOVATIONS IN ENGINEERING RESEARCH AND TECHNOLOGY [IJIERT]
ISSN: 2394-3696
VOLUME 2, ISSUE 7, JULY-2015
1 | P a g e
LPC Models and Different Speech Enhancement Techniques- A
Review
Mr. Arjun Ghule
Research Scholar, Department of ECE, J.N.T.U., Hyderabad, India
Dr.Prabhu Benakop
Director, Aurora`s Engineering College, Bongir, Hyderabad, India
Abstract
Author has already published one review paper on the quality enhancement of a speech signal by
minimizing the noise. This is a second paper of same series. In last two decades the researchers
have taken continuous efforts to reduce the noise signal from the speech signal. This paper
comments on, various study carried out and analysis proposals of the researchers for
enhancement of the quality of speech signal. Various models, coding, speech quality
improvement methods, speaker dependent codebooks, autocorrelation subtraction, speech
restoration, producing speech at low bit rates, compression and enhancement are the various
aspects of speech enhancement. We have presented the review of all above mentioned
technologies in this paper and also willing to examine few of the techniques in order to analyze
the factors affecting them in upcoming paper of the series.
Key words: LPC models, coding, speaker dependent codebooks, autocorrelation subtraction,
speech restoration, compression and enhancement of speech
Introduction
In last two decades, various models have been developed by the researchers for speech
enhancement. Many people have worked for analysis of the models and few others have
commented on the advantages and disadvantages of the models. We are writing the review of
some models developed and principles used while developing them. The source filter model of
speech is widely used for digital coding of speech; the popularity of the LPC method stems from
its computational efficiency in determining the synthesis filter parameters [2]. Speech
enhancement algorithms that reduce the level of noise present in a speech signal captured in the
presence of background noise have several applications [5].
Literature Review
Evangelos E. Milios and Alan V. Oppenheim [1] The work carried out by various researchers
has demonstrated the importance of the long-time phase of speech. The possibility of using the
long-time phase of the LPC residual signal in speech synthesis is investigated by author.
The importance of the various parts of the long-time spectrum of the residual is also
suggested. The potential for synthesizing high-quality speech using the phase-only residual has
been demonstrated. It is recommended that coding the long-time phase together with either the

ISSN: 2394-3696
2 | P a g e
smoothed long-time magnitude alone or the smoothed long-time magnitude augmented by some
more accurate low frequency magnitude information.
V. K. Jain, R. Hangartner [2] The introduction of multipulse excitation for LPC coders has
increased the quality achievable for digitally coded speech at bit rates in the 9.6 kbs range. A
simplified multipulse analysis is proposed with particular prominence on the speech model
developed as a result of a two pass method. In the first pass, estimated LPC parameters generated
by conventional covariance analysis are used to generate the forward prediction error; the
multi—pulse sequence is them detected by thresholding the residual. The second pass generates
the final LPC parameters by a covariance analysis incorporating knowledge of the estimated
pulse locations and amplitudes along with perceptual synthesis error weighting. Experimental
results are presented to demonstrate the method.
Rizwan Ishaq and Begofia Garcia Zapirain [3] The vocal fold modulates the air source from
lungs to produce voicing source for speech production. The vocal fold essential part of speech
production resides in larynx. The larynx cancer treatment necessitates removal of larynx, in
consequences normal speech production destroyed due to no voicing source available. The new
voicing source provided artificially or by use of Paryngo-esophageal (PE) segments. The voicing
source or residual signal for Esophageal (E) speech uses PE segment, has irregular behavior, and
produces degraded quality speech. This paper discussed and evaluated the residual signal or
voicing source enhancement of E speech by incorporating speech enhancement method Adaptive
Gain Equalizer (AGE), in time frequency and modulation frequency along with formants
modification by Line Spectral Frequencies (LSF) and Linear Predicative Coding (LPC). The
system validated by measuring Harmonic to Noise Ratio (HNR) temporally and maximum of
4dB enhancement has been observed in comparison to Kalman filtering based enhancement
where enhancement observed maximum of 2dB.
The improved HNR has validated the system capability to improve quality of E speech. The
system successfully removed noise, as well tried to enhanced residual signal for better and
intelligible E speech signal. The MAGE modulation frequency system provides better
enhancement in comparison to AGE and overall both systems outperformed Kalman filtering
based system [4]. Along with voicing source enhancement, formant enhancement significantly
improves quality of E speech in comparison to poles modification by shifting upward [4] when
conducted listening test. Although LPC analysis/synthesis provides good estimation of voicing
source signal, but as comparison to normal speech this estimation in not perfect and modeling of
PE segment can provides better voicing source signal which can be improved and modified
through this system. The future can be to provide optimized value of L opt by having noise
information from modeling of PE segment.
D. Hanumantha Rao Naidu, G. V. Prabhakara Rao, Sriram Srinivasan [5] Speech
enhancement techniques that employ trained codebooks of speech and noise linear predictive
coefficients have been previously shown to provide good performance in non stationary noise
environments. The speech codebooks are typically trained using data from a large number of
speakers. For noise reduction systems in personal devices such as mobile phones, the use of
speaker dependent codebooks is appealing as such devices are typically owned by a single user.

ISSN: 2394-3696
3 | P a g e
It is however not straightforward whether the use of speaker dependent codebooks translates into
an observable benefit in terms of noise reduction. Moreover, it is of interest to quantify the
improvement obtained over speaker independent codebooks. In this paper, the benefit of using
speaker dependent codebooks is analyzed in terms of the resulting improvement in noise
reduction when compared to using speaker independent codebooks.
The hypothesis that the use of speaker dependent (SD) codebooks can provide improved
performance compared to speaker independent (SI) codebooks when used for speech
enhancement has been validated through experimental results. It has been seen that the ability of
SD codebooks to better model the spectral shapes corresponding to a particular speaker's data
also translates into improvements in noise reduction using a codebook-based speech
enhancement method. SD codebooks provide a significant improvement over SI codebooks in
terms of all three performance measures considered in this paper: segmental signal-to-noise ratio,
log spectral distortion and PESQ scores, and indicate that the benefits justify the added
complexity in training the SD codebooks on-line. In fact, depending on the input SNR, it was
seen that an SD codebook of half the size of an SI codebook is sufficient to provide a similar
level of performance. Reducing the codebook size can significantly reduce the computational
complexity of the speech enhancement algorithm. This is desirable as the resulting savings more
than compensate for the added complexity in adapting the speech codebooks to the user's voice,
which only needs to be done when the user changes. Our results indicate that speaker dependent
approaches are promising for applications such as speech enhancement for personal devices, e.g.,
mobile phones. Future work will extend the investigation to gender and language dependent
codebooks. Another interesting extension is when the number of users is larger than one, but
limited to a small number. Examples of such applications include VoIP calls on a PC or home
telephony systems, where the users could be the members of a family. Future work will
investigate the use of one speech codebook per person, where the system automatically
determines the appropriate codebook to be used based on the input signal.
C. K. Mn and K. Y. Choi [6] A robust linear predictive coding (LPC) method that can be used
in noisy as well as quiet environment has been studied. In this method, noise autocorrelation
coefficients are first obtained and updated during non speech periods. Then, the effect of additive
noise in the input speech is removed by subtracting values of the noise auto correlation
coefficients from those of autocorrelation coefficients of corrupted speech in the course of
computation of linear prediction coefficients. When signal-to-noise ratio of the input speech
ranges from 0 to 10 dB, a performance improvement of about 5 dB can be gained by using this
method. The proposed method is computationally very efficient and requires a small storage
area. Authors have studied linear predictive coding in noisy environment and proposed a method
to reduced gradation caused by additive white noise. The approach is based on subtraction of
autocorrelation coefficients of noise from those of corrupted speech after estimation of noise
periodogram during intervals of non-speech activity. By using the proposed method, one can
improve the performance of an LPC vo coder by about 5 dB in SNR.
A.B. Premkumar, Ang Eeluang and A. S. Madhukumar[7] Restoration of speech parameters
that have been degraded by noisy channel is of paramount importance, since mutilated speech
parameters such as Linear Predictive Coefficients may render speech totally irrecoverable. The

ISSN: 2394-3696
4 | P a g e
objective of this paper is to provide a technique using discrete dyadic wavelet transforms as a
preprocessing stage prior to LPC extraction. We show that this step makes the speech parameters
more robust to noise perturbation in the channel and hence makes recovery of speech possible.
We discuss the effects of applying a single level and a three level transformation to speech prior
to compression and compare the quality of recovered speech in both cases. We also apply
Wiener filter to the wavelet coefficients and show that the recovered speech is of appreciable
quality.
Speech recovery from channel noise has been studied. Recovery has been made possible
using DWT techniques. Applying DWT analysis prior to the extraction of LPC parameters and
synthesizing speech from the inverse DWT coefficients has proved effective in combating
channel noise. This method works well so long as the formant information in speech is not
completely destroyed by noise affecting the LPC coefficients
Siva Sushma G, D R Sandeep [8] Over the past several years there has been considerable
attention focused on compression and enhancement of speech signals. This interest is progressed
towards the development of new techniques capable of producing good quality speech at the
output. Speech compression is a process of converting human speech into efficient encoded
representations that can be decoded to produce a close approximation of the original signal. In
this paper the simulated vo coder (LPC) using mat lab was implemented for compression. The
result obtained from LPC was compared with other implemented speech compression using
Wavelet transform in terms of quality. Speech enhancement means improving the quality or
value of the signal in a noisy environment. In this paper we proposed spectral subtraction (S.S)
and wavelet transform methods for de-noising and the result of one method was compared with
other. From the results we see that in both compression and enhancement, the performance of
wavelet transform was better than other.
In this paper, speech compression using LPC, Wavelets and enhancement using S.S, wavelet
transforms was presented. In both compression and enhancement of speech signals using wavelet
transforms, we have chosen DB20 wavelet and then the performance was tested and the
following points were observed. From the output signals of Lpc, we can observe that the noise
in the synthesized files is stronger than in the actual signals, but in wavelets the noisy feeling is
low. By this we can say that the signal to noise ratio (as one of the most important measurements
of the performance) of the Wavelet transform has high values comparable to LPC and high
compression is achieved using wavelet transforms. From the enhancement results, we can
observe that the signal to noise ratio of spectral subtraction is low when compared to wavelets.
By this we can say, that the spectral subtraction works poorly in de-noising when compared to
the wavelets. The analysis that we undertook for wavelets in both compression and enhancement
includes only the compactly supported wavelets. An improvement in the performance can be
done by studying the effects of other wavelets on compression and enhancement.
Bishnu S. Atal and Joel R. Remde [9] The excitation for LPC speech synthesis usually consists
of two separate signals - a delta-function pulse once every pitch period for voiced speech and
white noise for unvoiced speech. This manner of representing excitation requires that speech
segments be classified accurately into voiced and unvoiced categories and the pitch period of
voiced segments be known. it is now well recognized that such a rigid idealization of the vocal

ISSN: 2394-3696
5 | P a g e
excitation Li often responsible for the unnatural quality associated with synthesized speech. This
paper describes a new approach to the excitation problem that does not require a priori
knowledge of either the voiced-unvoiced decision or the pitch period. All classes of sounds are
generated by exciting the LPC filter with a sequence of pulses; the amplitudes and locations of
the pulses are determined using a non-iterative analysis-by-synthesis procedure. This procedure
minimizes a perceptual-distance metric representing subjectively-important differences between
the waveforms of the original and the synthetic speech signals. The distance metric takes account
of the finite-frequency resolution as well as the ditTerential sensitivity of the human ear to errors
in the formant and inter-formant regions of the speech spectrum.
Our work on developing a new model of LPC excitation for producing high quality
speech is as yet very preliminary. We find that speech quality can be significantly improved by
using multi pulse excitation signal. Moreover, the locations and amplitudes of the pulses in the
multi-pulse excitation can be determined by using a computationally efficient non-iterative
analysis-by synthesis procedure that can minimize the perceptual difference between the natural
and the synthetic speech waveforms. The difficult problems of voiced-unvoiced decision and
pitch analysis are eliminated.
Conclusion
Author has already published one review paper on the quality enhancement of a speech signal by
minimizing the noise. This is a second paper of same series. This paper comments on various
models and methodologies of the speech enhancement techniques. The various models proposed
by the scholars were working on various factors and the researchers also commented on the
methods for improving the performance of the models. All the models are good for various
surrounding conditions and there are several factors affecting the performance of each model.
We have presented the review of various mentioned technologies in this paper and also willing to
examine few of the techniques in order to analyze the factors affecting them in upcoming paper
of the series.
References
1] “The Phase-only Version of the LPC Residual in Speech Coding”- Evangelos E. Milios and
Alan V. Oppenheim, ICASSP 83, BOSTON
2] “Efficient Algorithm for multi—pulse LPC analysis of speech”, V. K. Jain, R. Hangartner,
ICASSP— 84
3] “Adaptive Gain Equalizer for Improvement of Esophageal Speech”, Rizwan Ishaq and
Begofia Garcia Zapirain, 978-1-4673-5604-6/12 ©20 12 IEEE.
4] B. Garcia and A. Mendez, "Oesophageal speech enhancement using poles stablization and
kalman filtering," ICASSP, pp. 1597-1600, 2008.

ISSN: 2394-3696
6 | P a g e
5] “Speech Enhancement Using Speaker Dependent Codebooks”, D. Hanumantha Rao Naidu, G.
V. Prabhakara Rao, Sriram Srinivasan, 978-1-4577-0274-7/11/©2011 IEEE, DSP 2011.
6] “Improving Lpc Nalysis Of Noisy Speech By Autocorrelation Subtraction Method”, C. K. Mn
and K. Y. Choi , CH161O-5/81/0000-1082 © 1981 IEEE
7] “Speech Restoration From Lpc Parameters Corrupted By Channel Noise”, A. B. Premkumar,
Ang Eeluang and A. S. Madhukumar, 1999 IEEE TENCON
8] “Compression and Enhancement of Speech Signals”, Siva Sushma G, D R Sandeep, Chennai
and Dr.MGR University Second International Conference on Sustainable Energy and Intelligent
System (SEISCON 2011) , Dr. M.G.R. University, Maduravoyal, Chennai, Tamil Nadu, India.
July. 20-22, 2011.
9] “A New Model Of Lpc Excitation For Producing Natural-Sounding Speech At Low Bit
Rates”, Bishnu S. Atal and Joel R. Remde, CH 1746-718210000 . 0614 © 1982 IEEE

LPC Models and Different Speech Enhancement Techniques- A Review

More Related Content

What's hot (19)

Similar to LPC Models and Different Speech Enhancement Techniques- A Review (20)

More from ijiert bestjournal (20)

Recently uploaded (20)

LPC Models and Different Speech Enhancement Techniques- A Review