Phoneme classification for speech synthesiser using differential EMG signals between muscles

Area of Science:

• Biomedical Engineering
• Signal Processing
• Speech Technology

Background:

• Traditional speech synthesis methods often require numerous electrodes for accurate phoneme classification.
• Electromyography (EMG) signals offer a potential avenue for non-invasive speech analysis.

Purpose of the Study:

• To propose and evaluate a novel method for phoneme classification using differential electromyography (EMG) signals between different muscles.
• To develop a Japanese speech synthesizer system that utilizes fewer electrodes through this new EMG-based approach.

Main Methods:

• Deriving EMG signals as the differential between monopolar signals from two distinct muscles, differing from traditional same-muscle bipolar derivations.
• Extracting frequency-based feature patterns using filter banks.
• Classifying phonemes with a probabilistic neural network that integrates feature reduction and pattern classification.

Main Results:

• The proposed differential EMG method achieved considerably high phoneme classification performance.
• The system demonstrated the feasibility of constructing a Japanese speech synthesizer with a reduced number of electrodes.

Conclusions:

• Differential EMG signals between muscles offer an effective and efficient approach for phoneme classification in speech synthesis.
• This technique significantly reduces the electrode count required for speech synthesizer systems, paving the way for more user-friendly and less invasive interfaces.