Voiceless Bangla vowel recognition using sEMG signal

Area of Science:

• Biomedical Engineering
• Speech Technology
• Signal Processing

Background:

• Vocal cord issues prevent sound production, complicating speech recognition for voiceless individuals.
• Surface Electromyogram (sEMG) offers a non-invasive alternative for capturing articulatory muscle activity.

Purpose of the Study:

• To develop a novel method for classifying Bangla vowels using sEMG signals.
• To enhance speech communication and voice synthesis for individuals with speech impairments.

Main Methods:

• Recorded sEMG signals during the pronunciation of eleven Bangla vowels.
• Applied mRMR feature selection to identify prominent features from extracted signal data.
• Utilized an Artificial Neural Network for vowel classification based on selected features.

Main Results:

• Achieved an overall classification accuracy of 82.3% for Bangla vowels.
• Demonstrated effective vowel recognition using a reduced feature set.
• Outperformed previous studies in different languages regarding feature efficiency.

Conclusions:

• The proposed sEMG-based method is effective for Bangla vowel classification in voiceless individuals.
• This technique offers a significant advancement for voice synthesis and speech communication technologies.
• The use of mRMR and Artificial Neural Networks provides a robust framework for bio-signal-based speech recognition.