Govind Divakaran
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Govind Divakaran has a total of 79 co-authors. Here are the top co-authors by number of shared publications:
D. Pravena
Amrita School of Engineering, Amrita Vishwa Vidyapeetham
10 shared publicationsS. R. Mahadeva Prasanna
Department Electronics and Electrical Engineering, Indian Institute of Technology Guwahati, Assam, India
4 shared publications- Akhilesh Kumar Dubey
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, India
3 shared publications - G. Jyothish Lal
Amrita Vishwa Vidyapeetham
3 shared publications - Gopalakrishnan Ennappadam Ananthanarayanan
Amrita School of Artificial Intelligence, Amrita Vishwa Vidyapeetham, Bengaluru, India
3 shared publications - S. R. Mahadeva Prasanna
Indian Institute of Technology Guwahati
3 shared publications
Latest Publications
- Open Access
- 15 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Characterizing stroke-affected speech using F0 and duration-based features
This paper presents a study of pitch (F0) and duration related features of transition and non-transition regions of speech, carried out to better understand and characterize speech recorded from stroke patients. A speech corpus consisting of read speech as well as five sustained vowels, recorded from 50 stroke patients and 50 healthy speakers in a clinical environment, was developed for this purpose. A gender specific statistical analysis of F0 contour showed that median F0 of an utterance shows consistent trend in distinguishing utterances from the stroke patients and healthy control speakers across all vowel categories. The utterances from female speakers of the stroke study group tended to have a lower F0 median in comparison with that of the healthy control group. In contrast, male speakers in the stroke study group exhibited a higher F0 median compared to the control group. The durations of the transition and non-transition (steady state) regions in the utterances were estimated by an analysis of the gradient of cepstral coefficient vector as a function of frame index. The male and female speakers of the stroke study group showed shorter transition regions with higher relative transition areas as compared to those of the healthy control group. In contrast, the non-transition (steady state) regions tend to show longer duration in the stroke study group. The observed statistical trends in the F0 and duration analysis were validated using one-way ANOVA tests.
- 15 February 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessCharacterizing stroke-affected speech using F0 and duration-based features
This paper presents a study of pitch (F0) and duration related features of transition and non-transition regions of speech, carried out to better understand and characterize speech recorded from stroke patients. A speech corpus consisting of read speech as well as five sustained vowels, recorded from 50 stroke patients and 50 healthy speakers in a clinical environment, was developed for this purpose. A gender specific statistical analysis of F0 contour showed that median F0 of an utterance shows consistent trend in distinguishing utterances from the stroke patients and healthy control speakers across all vowel categories. The utterances from female speakers of the stroke study group tended to have a lower F0 median in comparison with that of the healthy control group. In contrast, male speakers in the stroke study group exhibited a higher F0 median compared to the control group. The durations of the transition and non-transition (steady state) regions in the utterances were estimated by an analysis of the gradient of cepstral coefficient vector as a function of frame index. The male and female speakers of the stroke study group showed shorter transition regions with higher relative transition areas as compared to those of the healthy control group. In contrast, the non-transition (steady state) regions tend to show longer duration in the stroke study group. The observed statistical trends in the F0 and duration analysis were validated using one-way ANOVA tests.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 136751-136767
Evaluating Fundamental Frequency ( f 0 ) as an Acoustic Correlate of Stroke Severity From Speech
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 136751-136767
Open AccessEvaluating Fundamental Frequency ( f 0 ) as an Acoustic Correlate of Stroke Severity From Speech
- 29 October 2024
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 27 (4) , 1135-1147
Investigating the role of electroencephalogram based rhythmic bands: applications in imagined speech classification and epileptic seizure severity detection
- 29 October 2024
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 27 (4) , 1135-1147
Investigating the role of electroencephalogram based rhythmic bands: applications in imagined speech classification and epileptic seizure severity detection
- 8 August 2024
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 97
Variational mode decomposition based features for detection of hypernasality in cleft palate speech
- 8 August 2024
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 97
Variational mode decomposition based features for detection of hypernasality in cleft palate speech
- Open Access
- 9 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 98928-98944
Significance of Variational Mode Decomposition for Epoch Based Prosody Modification of Speech With Clipping Distortions
Clipping is one of the non-linear distortions commonly introduced due to microphone saturation during speech recording. Present work focuses on the effect of clipping in the task of prosody modification. Since, F 0 contour and duration are the important prosodic parameters, the present work studies the effect of clipping in the manipulation of F 0 and duration of a given speech. Epoch based prosody modification is considered as the popular method to generate waveforms with good perceptual quality by scaling F 0 contour and duration of the given speech by fixed scaling factors. Therefore, present work studies the effect of waveform clipping on the perceptual quality of prosody modified speech. Deviations in the estimation of epochs (which are used as the analysis pitch marks) and method used for generating the waveform are the two ways wherein perceptual quality in epoch based prosody modification can be compromised. The work proposed in this paper examines, effect of clipping on the aforesaid stages of epoch based prosody modification affecting the perceptual quality of the generated speech. Zero frequency filtering (ZFF), a simple and popular method, is chosen as the epoch estimation algorithm for epoch based prosody modification presented in the paper. Based on comparative epoch estimation performance analysis carried out by introducing various amplitude clipping levels, epoch identification rates are confirmed to be unchanged, irrespective of the level of clipping distortions present. However, due to saturation in the waveform samples, the waveform generation stage of the prosody modification was observed to be affected to the level which was proportional to the clipping distortions present in the signal. A variational mode decomposition (VMD) based signal approximation of the prosody modified speech is proposed to reduce the non-linear effect due to clipping. At the gross level, the re-estimated speech signal obtained from the VMD modes observed to have improved the perceptual quality of the pitch and duration modified speech. The improved perceptual quality of VMD based re-estimation of prosody modified speech was confirmed from subjective and NIST-STNR based objective assessments. Further, VMD based refinement is proposed as an alternative to local mean subtraction for trend removal in conventional ZFF of speech for the accurate epoch estimation. Comparative performance analysis carried out on CMU arctic database, confirmed improvement in the identification accuracy for the epochs estimated by using VMD based trend removal in ZFF algorithm.
- 9 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 98928-98944
Open AccessSignificance of Variational Mode Decomposition for Epoch Based Prosody Modification of Speech With Clipping Distortions
Clipping is one of the non-linear distortions commonly introduced due to microphone saturation during speech recording. Present work focuses on the effect of clipping in the task of prosody modification. Since, F 0 contour and duration are the important prosodic parameters, the present work studies the effect of clipping in the manipulation of F 0 and duration of a given speech. Epoch based prosody modification is considered as the popular method to generate waveforms with good perceptual quality by scaling F 0 contour and duration of the given speech by fixed scaling factors. Therefore, present work studies the effect of waveform clipping on the perceptual quality of prosody modified speech. Deviations in the estimation of epochs (which are used as the analysis pitch marks) and method used for generating the waveform are the two ways wherein perceptual quality in epoch based prosody modification can be compromised. The work proposed in this paper examines, effect of clipping on the aforesaid stages of epoch based prosody modification affecting the perceptual quality of the generated speech. Zero frequency filtering (ZFF), a simple and popular method, is chosen as the epoch estimation algorithm for epoch based prosody modification presented in the paper. Based on comparative epoch estimation performance analysis carried out by introducing various amplitude clipping levels, epoch identification rates are confirmed to be unchanged, irrespective of the level of clipping distortions present. However, due to saturation in the waveform samples, the waveform generation stage of the prosody modification was observed to be affected to the level which was proportional to the clipping distortions present in the signal. A variational mode decomposition (VMD) based signal approximation of the prosody modified speech is proposed to reduce the non-linear effect due to clipping. At the gross level, the re-estimated speech signal obtained from the VMD modes observed to have improved the perceptual quality of the pitch and duration modified speech. The improved perceptual quality of VMD based re-estimation of prosody modified speech was confirmed from subjective and NIST-STNR based objective assessments. Further, VMD based refinement is proposed as an alternative to local mean subtraction for trend removal in conventional ZFF of speech for the accurate epoch estimation. Comparative performance analysis carried out on CMU arctic database, confirmed improvement in the identification accuracy for the epochs estimated by using VMD based trend removal in ZFF algorithm.
- Open Access
- 15 February 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 57 (3) , 1-33
Milestones in speaker recognition
This article reviews significant research in the domain of speaker recognition, i.e., the task of determining the speaker’s identity from its speech. Unlike conventional review articles, this document strives to be concise and selective, provide a historical context, and reach a wider audience. In this endeavour, a summary of selected key works of every decade is provided which highlights the theme(s) of research of that period. At first, an overview of the humble beginnings of the 1960s and 70s is provided, followed by the key developments in the 80s and 90s. The prime focus of the research community in the 2000s is then discussed, leading to various non-conventional features, modelling techniques, and hybrid or fusion systems. The developments of the last decade (the 2010s), such as the i-vector-based systems, are then discussed. Modern speaker recognition based on Artificial Intelligence (AI), such as the x-vector system, and refinements of the i-vector-based systems using deep neural networks, are then discussed. The article concludes with a concise discussion of the evolving recent trends and allied research in speaker recognition.
- 15 February 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 57 (3) , 1-33
Open AccessMilestones in speaker recognition
This article reviews significant research in the domain of speaker recognition, i.e., the task of determining the speaker’s identity from its speech. Unlike conventional review articles, this document strives to be concise and selective, provide a historical context, and reach a wider audience. In this endeavour, a summary of selected key works of every decade is provided which highlights the theme(s) of research of that period. At first, an overview of the humble beginnings of the 1960s and 70s is provided, followed by the key developments in the 80s and 90s. The prime focus of the research community in the 2000s is then discussed, leading to various non-conventional features, modelling techniques, and hybrid or fusion systems. The developments of the last decade (the 2010s), such as the i-vector-based systems, are then discussed. Modern speaker recognition based on Artificial Intelligence (AI), such as the x-vector system, and refinements of the i-vector-based systems using deep neural networks, are then discussed. The article concludes with a concise discussion of the evolving recent trends and allied research in speaker recognition.
- 22 November 2023
- Published by Springer Nature
- p. 590-600
Significance of Duration Modification in Reducing Listening Effort of Slurred Speech from Patients with Traumatic Brain Injury
- 22 November 2023
- Published by Springer Nature
- p. 590-600
Significance of Duration Modification in Reducing Listening Effort of Slurred Speech from Patients with Traumatic Brain Injury
- 6 November 2023
- Published by Springer Nature in Arabian Journal of Geosciences
- Vol. 16 (12) , 1-21
Analysis and synthesis of East Himalayan earthquakes using a hybrid wavelet synchrosqueezed transform
- 6 November 2023
- Published by Springer Nature in Arabian Journal of Geosciences
- Vol. 16 (12) , 1-21
Analysis and synthesis of East Himalayan earthquakes using a hybrid wavelet synchrosqueezed transform
- 1 March 2023
- Published by SPIE-Intl Soc Optical Eng in Journal of Electronic Imaging
- Vol. 32 (02) , 023046-19
Automatic one-hand gesture (mudra) identification in bharatanatyam using eigenmudra projections and convolutional neural networks
Mudras in traditional Indian dance forms convey meaningful information when performed by an artist. The subtle changes between the different mudras in a dance form render automatic identification challenging as compared to conventional hand gesture identification, where the gestures are uniquely distinct from each other. Therefore, the objective of this study is to build a classifier model for the identification of the asamyukta mudra of bharatanatyam, one of the most popular classical dance forms in India. The first part of the paper provides a comprehensive review of the issues present in bharatanatyam mudra identification and the various studies conducted on the automatic classification of mudras. Based on this review, we observe that the unavailability of a large mudra corpus is a major challenge in mudra identification. Therefore, the second part of the paper focuses on the development of a relatively large database of mudra images consisting of 29 asamyukta mudras prevalent in bharatanatyam, which is obtained by incorporating different variabilities, such as subject, artist type (amateur or professional), and orientation. The mudra image database so developed is made available for academic research purposes. The final part of this paper describes the development of a convolutional neural network (CNN)-based automatic mudra identification system. Multistyle training of mudra classes on a conventional CNN showed a 92% correct identification rate. Based on the “eigenface” projection used in face recognition, “eigenmudras” projections of mudra images are proposed for improving the CNN-based mudra identification. Although the CNNs trained on the eigenmudra-projected images provide nearly equal identification rates as that obtained using the CNNs trained on raw mudra grayscale images, both models provide complementary mudra class information. The presence of complementary class information is confirmed by the improvement in the mudra identification performance when the CNN models trained from the raw mudra and eigenmudra-projected images are combined by computing the average of the scores obtained in the final softmax layers of both models. The same trend of improved mudra identification is observed upon combination of the average score level of VGG19 CNN models of the raw mudra images and corresponding eigenmudra-projected images.
- 1 March 2023
- Published by SPIE-Intl Soc Optical Eng in Journal of Electronic Imaging
- Vol. 32 (02) , 023046-19
Automatic one-hand gesture (mudra) identification in bharatanatyam using eigenmudra projections and convolutional neural networks
Mudras in traditional Indian dance forms convey meaningful information when performed by an artist. The subtle changes between the different mudras in a dance form render automatic identification challenging as compared to conventional hand gesture identification, where the gestures are uniquely distinct from each other. Therefore, the objective of this study is to build a classifier model for the identification of the asamyukta mudra of bharatanatyam, one of the most popular classical dance forms in India. The first part of the paper provides a comprehensive review of the issues present in bharatanatyam mudra identification and the various studies conducted on the automatic classification of mudras. Based on this review, we observe that the unavailability of a large mudra corpus is a major challenge in mudra identification. Therefore, the second part of the paper focuses on the development of a relatively large database of mudra images consisting of 29 asamyukta mudras prevalent in bharatanatyam, which is obtained by incorporating different variabilities, such as subject, artist type (amateur or professional), and orientation. The mudra image database so developed is made available for academic research purposes. The final part of this paper describes the development of a convolutional neural network (CNN)-based automatic mudra identification system. Multistyle training of mudra classes on a conventional CNN showed a 92% correct identification rate. Based on the “eigenface” projection used in face recognition, “eigenmudras” projections of mudra images are proposed for improving the CNN-based mudra identification. Although the CNNs trained on the eigenmudra-projected images provide nearly equal identification rates as that obtained using the CNNs trained on raw mudra grayscale images, both models provide complementary mudra class information. The presence of complementary class information is confirmed by the improvement in the mudra identification performance when the CNN models trained from the raw mudra and eigenmudra-projected images are combined by computing the average of the scores obtained in the final softmax layers of both models. The same trend of improved mudra identification is observed upon combination of the average score level of VGG19 CNN models of the raw mudra images and corresponding eigenmudra-projected images.