The FluentPlay Science 

AI & Signal Processing Logic

FluentPlay converts thought into speech support through real-time EEG decoding and intelligent signal processing. Our proprietary pipeline detects intention-to-speak states and fluency disruptions, delivering precision-timed cues that guide the user back to fluent control.

AI Classification Engine

Our AI model classifies brain states and predicts fluency outcomes in real-time.

• Architecture: Hybrid CNN + LSTM

• Output: Probability of fluent vs. blocked state

• Latency: <300 ms from signal to decision

Neural Signal Acquisition

FluentPlay uses the Emotiv EEG headset to capture high-fidelity brainwave activity from speech-related regions.

• Sampling Rate: 128 Hz – 256 Hz

• Signal Types: Delta, Theta, Alpha, Beta, Gamma

• Key Electrode Sites:

  • Frontal: speech initiation

  • Temporal: auditory feedback

  • Central: motor coordination

Feature Extraction

The cleaned signal is segmented and mined for fluency-relevant neural features:

• Power Spectral Density (PSD): mental effort & speech planning

• Event-Related Potentials (ERP): intention-to-speak detection

• Phase-Amplitude Coupling (PAC): timing of speech onset

Speech Synchronization Algorithm

When FluentPlay detects a fluent or blocked state, it initiates a cue to assist timing:

• Cue Types: Auditory, Haptic, or Visual

• Timing: ±150 ms alignment with cortical readiness

• Learning: Adaptive loop reduces cue reliance over time

Real-Time Preprocessing Pipeline

Each EEG stream is filtered, cleaned, and prepared within milliseconds to enable accurate interpretation:

• Noise Filtering

  • Bandpass: 0.5–40 Hz

  • Notch: 50/60 Hz powerline filter

• Artifact Rejection

  • ICA + supervised Machine Learning for eye blinks, motion, and muscle artifacts

• Normalization

  • Z-score standardization across channels and users