Journal of Scientific and Industrial Research

Open Access Open Access  Restricted Access Subscription Access

A Novel BCI - based Silent Speech Recognition using Hybrid Feature Extraction Techniques and Integrated Stacking Classifier


Affiliations
1 Department of Information Technology, PSG College of Technology, Coimbatore 641 004, Tamil Nadu, India
 

The Brain Computing Interface (BCI) is a technology that has resulted in the advancement of Neuro-Prosthetics applications. BCI establishes a connection between the brain and a computer system, primarily focusing on assisting, enhancing, or restoring human cognitive and sensory - motor functions. BCI technology enables the acquisition of Electroencephalography (EEG) signals from the human brain. This research concentrates on analyzing the articulatory aspects, including Wernicke's and Broca's areas, for Silent Speech Recognition. Silent Speech Interfaces (SSI) offers an alternative to conventional speech interfaces that rely on acoustic signals. Silent Speech refers to the process of communicating speech in the absence of audible and intelligible acoustic signals. The primary objective of this study is to propose a classifier model for phoneme classification. The input signal undergoes preprocessing, and feature extraction is carried out using traditional methods such as Mel Frequency Cepstrum Coefficients (MFCC), Mel Frequency Spectral Coefficients (MFSC), and Linear Predictive Coding (LPC). The selection of the best features is based on classification accuracy for a subject and is implemented using the Integrated Stacking Classifier. The Integrated Stacking Classifier outperforms other traditional classifiers, achieving an average accuracy of 75% for both thinking and speaking states on the KaraOne dataset and approximately 86.2% and 84.09% for thinking and speaking states on the Fourteen Channel EEG for Imagined Speech (FEIS) dataset.

Keywords

Electroencephalography, Linear predictive coding, Mel frequency cepstrum coefficients, Mel frequency spectral coefficients, Silent speech interface.
User
Notifications
Font Size

  • Sensinger J W & Dosen S, A review of sensory feedback in upper-limb prostheses from the perspective of human motor control, Front Neurosci, 14 (2020) 1–24, https://doi.org/ 10.3389/fnins.2020.00345.

  • Liu Q, Jiao Y, Miao Y, Zuo C, Wang X, Andrzej C & Jin J, Efficient representations of EEG signals for SSVEP frequency recognition based on deep multiset CCA, Neurocomputing, 378 (2020) 36–44, https://doi.org/10.1016/ j.neucom.2019.10.049.

  • Chuan-Chih H, Chia-Lung Y, Wai-Keung L, Hao-Teng H, Kuo-Kai S, Lieber Po-Hung Li , Tien-Yu Wui & Po-Lei Lee, Extraction of high-frequency SSVEP for BCI control using iterative filtering based empirical mode decomposition, Biomed Signal Process, 61 (2020) 1–12, https://doi.org/10.1016/j.bspc.2020.102022.

  • Wookey, Jessica, Busra, Bong, Azizbek & Suan, Biosignal sensors and Deep learning based speech Recognition – A Review, Sensors, 21 (2021) 1–22, https://doi.org/ 10.3390/s21041399.

  • Bhuvaneshwari M, Kanaga E G M & Anitha J, Bio-inspired Red Fox-Sine cosine optimization for the feature selection of SSVEP-based EEG signals for BCI applications, Biomed Signal Process Control, 80 (2022) 1–12, https://doi.org/10.1016/j.bspc.2022.104245.

  • Edla D R, Dodia S, Bablani A & Kuppili V, An efficient deep learning paradigm for deceit identification test on EEG Signals, ACM Trans Manag Inf Syst, 12 (2021) 1–20, https://doi.org/10.1145/3458791.

  • Mini P P, Tessamma T & Gopikakumari R, Wavelet feature selection of audio and imagined/vocalized EEG signals for ANN based multimodal ASR system, Biomed Signal Process Control, 63 (2021) 1–11, https://doi.org/10.1016/ j.bspc.2020.102218.

  • Sharon R A & Murthy H, Correlation based Multi-phasal models for improved imagined speech EEG recognition, Workshop on Speech, Music and Mind (2020) 21–25, https://doi.org/10.21437/smm.2020-5.

  • Clayton J, Wellington S, Valentini-Botinhao C & Watts O, Decoding imagined, heard and spoken speech: Classification and regression of EEG using a 14-channel dry-contact mobile headset (International speech communication Association) 2020, 4886–4890, https://doi.org/10.21437/ Interspeech.2020-2745.

  • Mansoor A, Usman M W, Jamil N & Naeem & M A, Deep learning Algorithm for Brain Computer Interface, Sci Programm, 2020 (2020) 1–12, doi:10.1155/2020/5762149.

  • Dash D, Ferrari P & Wang J, Decoding imagined and spoken phrases from non-invasive neural (MEG) signals, Front Neurosci, 14 (2020), doi: 10.3389/fnins.2020.00290.

  • Rusnac A L & Grigore O, Imaginary speech recognition using a convolutional network with long-short memory, Appl Sci, 12(22) (2022), 1–20, https://doi.org/10.3390/ app122211873.

  • Sharon R A, Narayanan S S, Sur M & Murthy A H, Neural Speech decoding during audition, imagination and production, IEEE Access, (2020), 149714–149727, doi: 10.1109/ACCESS.2020.3016756.

  • Saha P, Abdul-Mageed M & Fels S, Towards imagined speech recognition with hierarchical deep learning, arXiv: 1904.05746v1, (2019) 1–5.

  • Mahapatra N C & Bhuyan P, Multiclass classification of imagined speech vowels and words of electroencephalography signals using deep learning, Adv Hum Comput Interact, 2022 (2022) 1–10, doi:10.1155/2022/1374880.

  • Risanuri H, Agus B, Sujoko S & Anggun W, Denoising speech for MFCC feature extraction using wavelet transformation in speech recognition system, 10th Int Conf Info Technol Electr Eng (IEEE) 2018, 1–5, doi: 10.1109/ICITEED.2018.8534807.

  • Gupta H & Gupta D, LPC and LPCC method of feature extraction in speech recognition system, Int Conf Cloud System Big data Eng (IEEE) 2016, 1–5, doi: 10.1109/confluence.2016.7508171.

  • Ahmad J & Hayat M, MFSC: Multi-voting based feature selection for classification of Golgi proteins by adopting the general form of Chou's PseAAC components, J Theor Biol, 463 (2019) 1–29, https://doi.org/10.1016/ j.jtbi.2018.12.017


Abstract Views: 116

PDF Views: 65




  • A Novel BCI - based Silent Speech Recognition using Hybrid Feature Extraction Techniques and Integrated Stacking Classifier

Abstract Views: 116  |  PDF Views: 65

Authors

N. Ramkumar
Department of Information Technology, PSG College of Technology, Coimbatore 641 004, Tamil Nadu, India
D. Karthika Renuka
Department of Information Technology, PSG College of Technology, Coimbatore 641 004, Tamil Nadu, India

Abstract


The Brain Computing Interface (BCI) is a technology that has resulted in the advancement of Neuro-Prosthetics applications. BCI establishes a connection between the brain and a computer system, primarily focusing on assisting, enhancing, or restoring human cognitive and sensory - motor functions. BCI technology enables the acquisition of Electroencephalography (EEG) signals from the human brain. This research concentrates on analyzing the articulatory aspects, including Wernicke's and Broca's areas, for Silent Speech Recognition. Silent Speech Interfaces (SSI) offers an alternative to conventional speech interfaces that rely on acoustic signals. Silent Speech refers to the process of communicating speech in the absence of audible and intelligible acoustic signals. The primary objective of this study is to propose a classifier model for phoneme classification. The input signal undergoes preprocessing, and feature extraction is carried out using traditional methods such as Mel Frequency Cepstrum Coefficients (MFCC), Mel Frequency Spectral Coefficients (MFSC), and Linear Predictive Coding (LPC). The selection of the best features is based on classification accuracy for a subject and is implemented using the Integrated Stacking Classifier. The Integrated Stacking Classifier outperforms other traditional classifiers, achieving an average accuracy of 75% for both thinking and speaking states on the KaraOne dataset and approximately 86.2% and 84.09% for thinking and speaking states on the Fourteen Channel EEG for Imagined Speech (FEIS) dataset.

Keywords


Electroencephalography, Linear predictive coding, Mel frequency cepstrum coefficients, Mel frequency spectral coefficients, Silent speech interface.

References