Inner Speech Decoding: A Comprehensive Review.

Area of Science:

• Neuroscience and Computer Science, focusing on machine learning applications in brain-computer interfaces.

Background:

• Inner speech decoding aims to interpret silent speech from neural signals, offering communication alternatives for individuals with disabilities.
• Existing research explores various recording modalities and machine learning strategies to infer covert speech.

Purpose of the Study:

• To review current inner speech decoding technologies, including recording methods and machine learning algorithms.
• To discuss the factors influencing decoding accuracy, such as vocabulary size, sensor quality, and data.
• To examine the ethical, societal, and regulatory challenges associated with brain data privacy and user trust.

Main Methods:

• Review of non-invasive (EEG) and invasive (electrocorticography) recording modalities.
• Analysis of machine learning strategies including linear discriminant analysis, deep convolutional networks, and EEG-fMRI fusion.
• Focus on public datasets of imagined or articulated speech to illustrate impacts of sensor quality, data size, and domain adaptation.

Main Results:

• Small vocabularies under controlled conditions can achieve reasonable accuracy in inner speech decoding.
• Context-based approaches can further refine decoding outcomes.
• Effective inner speech recognition requires high-quality preprocessing, subject calibration, and informed modeling choices.

Conclusions:

• Inner speech decoding technology holds significant potential for communication assistance in impaired populations.
• Methodological standards, including preprocessing and calibration, are crucial for reliable decoding.
• Addressing ethical and societal challenges is vital for responsible development and adoption of this technology.