SlideShare a Scribd company logo
Silent Speech Recognition
Array Based Electromyographic Silent Speech Interface
Catalogue
• Introduction

• Electromyograph & Electromyography
• Working of Electromyograph
• Experiment for silent speech interface working

• Results of the Experiment
• Conclusion
• References

2
Introduction
• Discussing Speech recognition based on electric potentials of human articulatory
muscles.
• Using an EMG recording system based on multi-channel electrode arrays.
• Firstly we discuss what is Electromygraphy and then we discuss as experiment
conducted along with its results.

3
Electromyography
• Electromyography (EMG) is a technique for evaluating and recording the
electrical activity produced by skeletal muscles. EMG is performed using an
instrument called an electromyograph, to produce a record called an
electromyogram.
• An electromyograph detects the electrical potential generated by muscle cells
when these cells are electrically or neurologically activated.

• The signals can be analyzed to detect medical abnormalities, activation level,
recruitment order or to analyze the biomechanics of human or animal movement.

4
Working of Electromyograph
• When a muscle fiber contracts, small electrical currents in form of ion flows are
generated.
• EMG electrodes attached to the subject’s face capture the potential differences
arising from these ion flows. This allows speech to be recognized even when it is
produced silently, i.e. mouthed without any vocal effort.
• So far, all EMG-based speech recognizers have re- lied on small sets of less than
10 EMG electrodes attached to the speaker’s face.

5
Electromyograph

6
Experiment
• In setup A, we uni-polarly recorded 16 EMG channels with two EMG arrays each
featuring a single row of 8 electrodes, with 5 mm inter-electrode distance (IED).
One of the arrays was attached to the subject’s cheek, capturing several major
articulatory muscles (Maier-Hein et al., 2005), the other one was attached to the
subject’s chin, in particular recording signals from the tongue. A reference
electrode was placed on the subject’s neck.
• In setup B, we replaced the cheek array with a larger array containing four rows
of 8 electrodes, with 10 mm IED. The chin array remained in its place. In this
setup, we achieved a cleaner signal by using a bipolar configuration, where the
potential difference between two adjacent channels in a row is measured.

7
Experiment

(Cont.)

• We have four setups to investigate, namely, setups A-1 and A-2 (with 16 EMG
channels) and B-1 and B-2 (with 35 EMG channels).
• At this point we remark that the results on the four setups are not directly
comparable, since the number of training sentences, the set of speakers and the
number of sessions per speaker differ.
• The following table summarizes the properties of the corpus used:

8
Experiment

(Cont.)

9
Experiment Results
• Our first experiment establishes a baseline recognition system. We use our
recognizer, as described in section 3, and feed it with the EMG features from the
array recording system. Figure 2 shows the Word Error Rates for different
stacking widths, averaged over all sessions of the four setups.
• Our observations for the four distinct setups presented in this study are very
different: For setup A-1, with 16 channels and 40 training sentences, the Word
Error Rate (WER) varies between 46.3% and 53.2%, with the optimum reached at
a context width of 5 (i.e. TD5). For the B-1 setup, with 35 channels but the same
amount of training data, the optimal con- text width appears to be TD2 with a
WER of 50.5%, and for wider contexts, which increases to 87.6% for the TD15
stacking.

10
Experiment Results

(Cont.)

• For the setups with 160 training sentences, the recognition performance is
generally better due to the increased training data amount. With respect to context widths, we observe a behaviour which vastly differs from the results above:
For 16 EMG channels (setup A-2), the optimal context width is TD15, with a WER
of only 17.0%. For setup B-2, TD5 stacking is optimal, with a WER of 13.4%.

11
Conclusion
• In this study we have laid the basics of a new EMG-based speech recognition
technology, based on electrode arrays instead of single electrodes. We have
presented two basic recognition setups and evaluated their potential on data sets
of different sizes. The unexpected inconsistency with respect to the optimal
stacking width led us to the introduction of a PCA pre-processing step before the
LDA matrix is computed, which gives us consistent relative Word Error Rate
improvements of 10% to 18%, even for small training data sets of only 40
sentences.

12
References
• http://en.wikipedia.org/wiki/Electromyography

• Electromyographic Silent Speech Interface - Michael Wand, Christopher
Schulte, Matthias Janke, Tanja Schultz (Cognitive Systems Lab, Karlsruhe
Institute of Technology, Karlsruhe, Germany)

13

More Related Content

PPTX
Silent Sound Technology
PDF
Silent sound technology final report
DOCX
Abstract Silent Sound Technology
PPTX
Silent sound technology
PPTX
Silent sound technology
PPT
Silent sound technology
PPTX
Silent Sound Technology
PPTX
Silent sound technology_powerpoint
Silent Sound Technology
Silent sound technology final report
Abstract Silent Sound Technology
Silent sound technology
Silent sound technology
Silent sound technology
Silent Sound Technology
Silent sound technology_powerpoint

What's hot (20)

PDF
Silent Sound Technology
PPTX
Silent sound technology- Technology towards change.
DOCX
Silent Sound Technology
PDF
silent sound technology final report(17321A0432) (1).pdf
PPT
silent sound technology
PPTX
Silent sound technology
PPTX
Silent sound technology
PPTX
Silent sound technology NEW
PPTX
Silent Sound Technology
PPTX
Silent sound technologyrevathippt
DOCX
Automatic Speech Recognition
PPTX
SILENT SOUND TECHNOLOGY
PDF
INTERNET OF BEHAVIOUR (IOB).pdf
PPTX
SILENT SOUND TECHNOLOGY ppt.pptx
PPT
Silent sound-technology ppt final
PPTX
Speech to text conversion
PPT
Automatic number plate recognition
DOCX
Speech Recognition by Iqbal
PPT
Automatic speech recognition
PDF
speech processing and recognition basic in data mining
Silent Sound Technology
Silent sound technology- Technology towards change.
Silent Sound Technology
silent sound technology final report(17321A0432) (1).pdf
silent sound technology
Silent sound technology
Silent sound technology
Silent sound technology NEW
Silent Sound Technology
Silent sound technologyrevathippt
Automatic Speech Recognition
SILENT SOUND TECHNOLOGY
INTERNET OF BEHAVIOUR (IOB).pdf
SILENT SOUND TECHNOLOGY ppt.pptx
Silent sound-technology ppt final
Speech to text conversion
Automatic number plate recognition
Speech Recognition by Iqbal
Automatic speech recognition
speech processing and recognition basic in data mining
Ad

Viewers also liked (6)

PPT
Silent Speech
PPT
Ear recognition system
PPT
Ear Biometrics
PPTX
Silent sound interface
PDF
Silentsound documentation
PPTX
Deep neural networks
Silent Speech
Ear recognition system
Ear Biometrics
Silent sound interface
Silentsound documentation
Deep neural networks
Ad

Similar to Silent speech recognition (20)

PDF
PDF
06972937
PDF
018 bci gamma band
PDF
Suitable Mother Wavelet Selection for EEG Signals Analysis: Frequency Bands D...
PDF
SUITABLE MOTHER WAVELET SELECTION FOR EEG SIGNALS ANALYSIS: FREQUENCY BANDS D...
PDF
Improved Algorithm for Brain Signal Analysis
PDF
Improved Algorithm for Brain Signal Analysis
PPTX
Ffeature extraction of epilepsy eeg using discrete wavelet transform
PPT
Eeg examples
PDF
English For The Students Of Biomedical Engineering Siamk Najarian
PPTX
Basics of eeg signal
PDF
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
PDF
⭐⭐⭐⭐⭐ EMG Signal Processing with Clustering Algorithms for Motor Gesture Tasks
PPTX
Electromyography (EMG)
PDF
A Detail Study of Wavelet Families for EMG Pattern Recognition
PDF
A comparative study of wavelet families for electromyography signal classific...
PDF
AnalysisofEEGsignal_for_education___.pdf
PPTX
ecochG.pptx
PDF
F3602045049
PPT
A Novel Single-Trial Analysis Scheme for Characterizing the Presaccadic Brain...
06972937
018 bci gamma band
Suitable Mother Wavelet Selection for EEG Signals Analysis: Frequency Bands D...
SUITABLE MOTHER WAVELET SELECTION FOR EEG SIGNALS ANALYSIS: FREQUENCY BANDS D...
Improved Algorithm for Brain Signal Analysis
Improved Algorithm for Brain Signal Analysis
Ffeature extraction of epilepsy eeg using discrete wavelet transform
Eeg examples
English For The Students Of Biomedical Engineering Siamk Najarian
Basics of eeg signal
Analysis of EEG data Using ICA and Algorithm Development for Energy Comparison
⭐⭐⭐⭐⭐ EMG Signal Processing with Clustering Algorithms for Motor Gesture Tasks
Electromyography (EMG)
A Detail Study of Wavelet Families for EMG Pattern Recognition
A comparative study of wavelet families for electromyography signal classific...
AnalysisofEEGsignal_for_education___.pdf
ecochG.pptx
F3602045049
A Novel Single-Trial Analysis Scheme for Characterizing the Presaccadic Brain...

Recently uploaded (20)

PPTX
A Presentation on Touch Screen Technology
PDF
1 - Historical Antecedents, Social Consideration.pdf
PPTX
TLE Review Electricity (Electricity).pptx
PDF
Hybrid model detection and classification of lung cancer
PDF
Profit Center Accounting in SAP S/4HANA, S4F28 Col11
PPTX
Group 1 Presentation -Planning and Decision Making .pptx
PDF
NewMind AI Weekly Chronicles - August'25-Week II
PDF
DP Operators-handbook-extract for the Mautical Institute
PPTX
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
PPTX
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
PDF
Web App vs Mobile App What Should You Build First.pdf
PDF
Accuracy of neural networks in brain wave diagnosis of schizophrenia
PDF
From MVP to Full-Scale Product A Startup’s Software Journey.pdf
PPTX
1. Introduction to Computer Programming.pptx
PDF
DASA ADMISSION 2024_FirstRound_FirstRank_LastRank.pdf
PDF
A comparative analysis of optical character recognition models for extracting...
PDF
project resource management chapter-09.pdf
PPTX
Chapter 5: Probability Theory and Statistics
PDF
Hindi spoken digit analysis for native and non-native speakers
PDF
Assigned Numbers - 2025 - Bluetooth® Document
A Presentation on Touch Screen Technology
1 - Historical Antecedents, Social Consideration.pdf
TLE Review Electricity (Electricity).pptx
Hybrid model detection and classification of lung cancer
Profit Center Accounting in SAP S/4HANA, S4F28 Col11
Group 1 Presentation -Planning and Decision Making .pptx
NewMind AI Weekly Chronicles - August'25-Week II
DP Operators-handbook-extract for the Mautical Institute
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
Web App vs Mobile App What Should You Build First.pdf
Accuracy of neural networks in brain wave diagnosis of schizophrenia
From MVP to Full-Scale Product A Startup’s Software Journey.pdf
1. Introduction to Computer Programming.pptx
DASA ADMISSION 2024_FirstRound_FirstRank_LastRank.pdf
A comparative analysis of optical character recognition models for extracting...
project resource management chapter-09.pdf
Chapter 5: Probability Theory and Statistics
Hindi spoken digit analysis for native and non-native speakers
Assigned Numbers - 2025 - Bluetooth® Document

Silent speech recognition

  • 1. Silent Speech Recognition Array Based Electromyographic Silent Speech Interface
  • 2. Catalogue • Introduction • Electromyograph & Electromyography • Working of Electromyograph • Experiment for silent speech interface working • Results of the Experiment • Conclusion • References 2
  • 3. Introduction • Discussing Speech recognition based on electric potentials of human articulatory muscles. • Using an EMG recording system based on multi-channel electrode arrays. • Firstly we discuss what is Electromygraphy and then we discuss as experiment conducted along with its results. 3
  • 4. Electromyography • Electromyography (EMG) is a technique for evaluating and recording the electrical activity produced by skeletal muscles. EMG is performed using an instrument called an electromyograph, to produce a record called an electromyogram. • An electromyograph detects the electrical potential generated by muscle cells when these cells are electrically or neurologically activated. • The signals can be analyzed to detect medical abnormalities, activation level, recruitment order or to analyze the biomechanics of human or animal movement. 4
  • 5. Working of Electromyograph • When a muscle fiber contracts, small electrical currents in form of ion flows are generated. • EMG electrodes attached to the subject’s face capture the potential differences arising from these ion flows. This allows speech to be recognized even when it is produced silently, i.e. mouthed without any vocal effort. • So far, all EMG-based speech recognizers have re- lied on small sets of less than 10 EMG electrodes attached to the speaker’s face. 5
  • 7. Experiment • In setup A, we uni-polarly recorded 16 EMG channels with two EMG arrays each featuring a single row of 8 electrodes, with 5 mm inter-electrode distance (IED). One of the arrays was attached to the subject’s cheek, capturing several major articulatory muscles (Maier-Hein et al., 2005), the other one was attached to the subject’s chin, in particular recording signals from the tongue. A reference electrode was placed on the subject’s neck. • In setup B, we replaced the cheek array with a larger array containing four rows of 8 electrodes, with 10 mm IED. The chin array remained in its place. In this setup, we achieved a cleaner signal by using a bipolar configuration, where the potential difference between two adjacent channels in a row is measured. 7
  • 8. Experiment (Cont.) • We have four setups to investigate, namely, setups A-1 and A-2 (with 16 EMG channels) and B-1 and B-2 (with 35 EMG channels). • At this point we remark that the results on the four setups are not directly comparable, since the number of training sentences, the set of speakers and the number of sessions per speaker differ. • The following table summarizes the properties of the corpus used: 8
  • 10. Experiment Results • Our first experiment establishes a baseline recognition system. We use our recognizer, as described in section 3, and feed it with the EMG features from the array recording system. Figure 2 shows the Word Error Rates for different stacking widths, averaged over all sessions of the four setups. • Our observations for the four distinct setups presented in this study are very different: For setup A-1, with 16 channels and 40 training sentences, the Word Error Rate (WER) varies between 46.3% and 53.2%, with the optimum reached at a context width of 5 (i.e. TD5). For the B-1 setup, with 35 channels but the same amount of training data, the optimal con- text width appears to be TD2 with a WER of 50.5%, and for wider contexts, which increases to 87.6% for the TD15 stacking. 10
  • 11. Experiment Results (Cont.) • For the setups with 160 training sentences, the recognition performance is generally better due to the increased training data amount. With respect to context widths, we observe a behaviour which vastly differs from the results above: For 16 EMG channels (setup A-2), the optimal context width is TD15, with a WER of only 17.0%. For setup B-2, TD5 stacking is optimal, with a WER of 13.4%. 11
  • 12. Conclusion • In this study we have laid the basics of a new EMG-based speech recognition technology, based on electrode arrays instead of single electrodes. We have presented two basic recognition setups and evaluated their potential on data sets of different sizes. The unexpected inconsistency with respect to the optimal stacking width led us to the introduction of a PCA pre-processing step before the LDA matrix is computed, which gives us consistent relative Word Error Rate improvements of 10% to 18%, even for small training data sets of only 40 sentences. 12
  • 13. References • http://en.wikipedia.org/wiki/Electromyography • Electromyographic Silent Speech Interface - Michael Wand, Christopher Schulte, Matthias Janke, Tanja Schultz (Cognitive Systems Lab, Karlsruhe Institute of Technology, Karlsruhe, Germany) 13