WinterLightLabs careers

We have created a robust processing pipeline that extracts over 400 variables from a short sample of speech, both from the acoustics of a person’s voice and the automatically generated transcript of the words being communicated. Voice acoustics include measures of pause rate, duration and type, as well as features imperceptible to humans, such as spectral coefficients related to articulator motion. Lexicosyntactic variables assess the types of words used and the complexity of sentences, while semantic measures quantify information content. Combining these features with machine learning algorithms allows us to distinguish between healthy and impaired speech patterns.