Lyre is a machine learning–powered tool designed to detect deception in text, audio, and video. It analyzes patterns in language, speech, and behavior to identify signals that may indicate dishonesty. It learns from large datasets—spotting subtle patterns in sentence structure, voice inflection and facial expression that correlate with not being truthful.
Yes.
It is similar to polygraph, in that it detects human signals correlated with lying.
Unlike polygraphs which measure physiological signals in real time, requiring elaborate equipment, a trained operator and the physical presence of the subject, Lyre uses machine learning to analyze signals in text, audio or video files to flag behavior correlated with deception.
It can detect these patterns to a much greater degree of subtlety than any human, can process these files thousands of times faster than polygraph, and with higher accuracy without a trained operator or the physical presence of the subject.
Neither method is a fact checker. They both flag human signals, but require a holistic judgment from the operator on whether deception has occurred.
Lyre 2.0 achieves over 85% accuracy by analyzing text, audio, and video together. Its multi-modal AI model detects deception more reliably than single-source tools by combining language patterns, vocal cues, and facial expressions for a fuller picture.
Lyre uses bias mitigation techniques during model training to reduce gender disparities in outcomes. We apply a weighting strategy informed by fairness metrics, developed using IBM’s AI Fairness 360 (AIF360) toolkit. This helps ensure the model treats responses from different gender groups more equitably, improving fairness without compromising accuracy.
Lyre is trained on English language data. Different language models are achievable in Lyre’s design and part of our future product development. Alternative language training requires a suitable test data set and can be done in a matter of weeks.
Lyre is compatible with most common text, audio, and video files types. It can analyze typed documents, transcribed conversations, recorded speech, or video interviews for signs of deception.
Yes. Lyre offers flexible API integration, allowing it to be embedded directly into your internal tools, platforms, or customer-facing products.
Lyre is fast enough to be used in near real time for text and audio. Specific performance depends on data file type and hardware power, and other implementation details.
No.
Lyre is a decision-support tool to make analysis faster. It surfaces signals that may indicate deception, but human judgment is always essential for interpretation and context.