Lpc analysis of kannada speech

1. “LPC Analysis of Kannada Syllables”
Under the Guidance of
Prof. K. Indira
Dept. of E&C, RIT, Bangalore
Presented by
Vinodkumar A G - 1MS20EC127
Sridhar B - 1MS20EC111
Vinayak M B - 1MS20EC125
Vinay Swastik H - 1MS20EC124
MAJOR PROJECT

2. Table of Contents
1. Introduction and overview of the project
2. Problem statement
3. Project objectives and scope
4. Literature survey
5. Methodology, Proposed Work and Preliminary Results
6. References

3. INTRODUCTION
• The project will involve implementing the LPC algorithm to model the vocal tract, extracting the
formant frequencies from the LPC model, and comparing the results with known formant frequencies.
• The estimation of formant frequencies is an important task in speech signal processing, as it provides
information about the spectral characteristics of the vocal tract. Linear Predictive Coding (LPC) is a
widely used technique for speech analysis and has been shown to be effective for estimating formant
frequencies.
• Overall, the goal of this project is to develop a practical understanding of LPC and its applications in
speech analysis and to gain insights into the spectral characteristics of the vocal tract.

4. PROBLEM STATEMENT
To estimate Formant frequencies of the Kannada
Vowels and Consonants using Linear Predictive
Coefficients and verify using Praat tool.

5. Project Objectives
 The main objective of this report is to estimate the formant frequencies of Kannada
vowels and consonants using Linear Predictive Coding (LPC) analysis.
 The formant frequencies of speech sounds provide information about the resonant
properties of the vocal tract, which are essential for understanding the acoustic properties
of speech sounds.
 The estimated formant frequencies can be used for further analysis and modeling of
Kannada speech signals.
 An additional objective may involve applying the estimated formant and pitch
frequencies in practical speech processing tasks. This could include applications such as
speech recognition, speech synthesis, or speaker identification specific to Kannada
language.

6. Literature Survey
1.Performance Analysis of Kannada Phonetics: Vowels, Fricatives and Stop
Consonants Using LP Spectrum.
-By Shivakumar M and Latha Mariswamy
• A dataset of Kannada speech samples is collected, including a variety of phonetic units such as vowels,
fricatives, and stop consonants, recorded from native Kannada speakers.
• LP spectrum analysis is performed on the collected speech samples to estimate the spectral envelope and
formant frequencies.
• The study presents the findings of the spectral analysis, highlighting the spectral characteristics, patterns,
and variations observed in Kannada vowels, fricatives, and stop consonants.
• The accuracy and effectiveness of the LP spectrum analysis in capturing the phonetic properties of Kannada
are evaluated by comparing the estimated spectral features with known phonetic characteristics.

7. 2.Formants and LPC Analysis of Kannada Vowels Speech Signal
-By K. Indira, Sadashiva Chakrasali and Umesh Bilembagi
• The speech signal is down-sampled by a factor of 6 after passing through a low-pass filter, resulting
in a sampling frequency of 7350Hz.
• Pre-emphasis is applied to enhance the power of high frequency signals before LPC coefficients are
extracted using an autoregressive filter of varying order.
• The LPC filter is then used to obtain the LP residual, and frequency responses of LPC filters for
different orders are compared with the formants of corresponding vowels noted from a tool.

8. 3.Extraction of Speech Pitch and Formant Frequencies using
Discrete Wavelet Transform.
-By Sajad Hamzenejadi, Seyed Amir Yousef Hosseini Goki and Mahdieh Ghazvini
• The paper proposes a method for estimating speech pitch and formant frequencies using Discrete
Wavelet
Transform.
• DWT is used to decompose the speech signal into sub-bands, and the pitch and formant frequencies are
estimated from each sub-band.
• The method is advantageous because it captures both time and frequency information, and is efficiently
implemented using filter banks.
• The proposed method is shown to outperform existing methods for pitch and formant frequency
estimation in terms of accuracy and robustness.

9. 4.Formant Text to Speech Synthesis Using Artificial Neural Networks.
-By Gurinder Kaur and Parminder Singh
• The paper proposes a method for formant-based Text-to-Speech (TTS)
synthesis using Artificial Neural Networks (ANN).
• The method involves training an ANN on a set of formant frequency
parameters and their corresponding phonetic labels to generate synthetic
speech.
• The paper discusses the advantages of using formant-based synthesis over
concatenative TTS, including improved naturalness and flexibility.
• The proposed method is shown to achieve high-quality speech synthesis with
low computational complexity and outperform existing methods in terms of
naturalness and intelligibility.

10. Methodology:
• Block Diagram:

11. Steps:
1. Collection of Speech Samples.
2. Pre-processing
3. Frame the speech signal
4. Compute LPC coefficients
5. Compute formant frequencies
6. Inverse filtering
7. Comparing formants of MATLAB with Praat Tool formants

12. Letter: ಅ(Male(23))

13. Letter: ಆ (Male(23))

14. Letter : ಅ(Female(18))

15. Letter : ಆ (Female(18))

16. Formant Frequencies of Male person ( Age 20-25): Vowels

17. Formant Frequencies of Male person ( Age 20-25): Consonants

18. Conclusion:
• In this work Kannada vowels and consonants were recorded from different age groups.
• Formants frequencies of corresponding Vowels and Consonants were computed. The variation of
formant frequencies across different gender and different age groups are shown in tables.
• The analysis is carried out separately for male and female speakers. The preliminary analysis of
frequency domain characteristics of vowels shows significant variations across different gender
and age groups.
• The importance of F1, F2, F3, F4 (formants) and their impact on order of the LPC filter have been
studied thoroughly in great details. The results have indicated the significant dependency of
speech signal characteristics on gender and different age groups.

19. Future work:
1.LPC-based formant frequency estimation can be used for speech enhancement,
speech recognition, voice conversion, and speech pathology diagnosis.
2.In speech enhancement, LPC can help to remove noise and other unwanted
distortions from speech signals.
3.In speech recognition, LPC-based formant frequency estimation can improve
accuracy and mitigate the effects of noise and other distortions.
4.In voice conversion, LPC can be used to convert the formant frequencies of one
speaker's voice to those of another speaker.
5.In speech pathology diagnosis, LPC-based formant frequency estimation can be
used to identify deviations from normal speech patterns and assist in diagnosis.

20. REFERENCES
1. Latha, M., M. Shivakumar, and R. Manjula. "Performance Analysis of Kannada Phonetics: Vowels,
Fricatives and Stop Consonants Using LP Spectrum." SN Computer Science 1, no. 2 (2020): 84.
2. Chakrasali Sadashiva, Umesh Bilembagi, and K. Indira. "Formants and LPC analysis of
Kannada vowel speech signals." In 2018 3rd IEEE International Conference on Recent
Trends in Electronics,Information & Communication Technology (RTEICT), pp. 945-948.
IEEE, 2018.
3. Dhiman Chowdhury , Md. Raju Ahmed Ripan,Md. Mehedihasan “Speech Features:
Pitch and Formant Extraction of Vowel Sounds Using Autocorrelation and Frequency
Domain Spectral Analysis” International conference on Innovation in Engineering
and Technology (ICIET),27-29 Dec 2018.
4. Sajad Hamzenejad , Seyed Amir Yousef Hosseini Goki, Mahdieh Ghazvini “Extraction
of Speech Pitch and Formant Frequencies using Discrete Wavelet Transform”, 2019 7th
Iranian Joint Congress on Fuzzy and Intelligent Systems (CFIS).

21. 5. U. Shrawankar and V. Thakare, “Feature Extraction for a Speech Recognition System in Noisy
Environment: A Study”, in Proc. Second Int. Conf. on Computer Engineering and Applications, 19-21
Mar. 2010.
6. EV Raghavendra, P. Vijay Aditya and K. Prahalad, "Speech synthesis using artificial neural networks
2010 National Conference On Communications (NCC) Chennai, India 2010, pp: 1-5, dor 10.1109/NCC
2010 5430190
7. Reddy MV, Hanumanthappa M. Kannada phonemes to speech dictionary: statistical approach. Int J Eng
Res Appl. 2017;7(1):77–80.
8. Sarika Hegde KK, Achary KK, Shetty S. Statistical analysis of features and classification of alpha
syllabary, sounds in Kannada language. New York: Springer; 2014.
9. Formant Text To Speech Synthesis Using Artificial Neural Networks, 2019 Second International
Conference on Advanced Computational and Communication Paradigms (ICACCP).

22. This Photo by Unknown author is licensed under CC BY-SA.