AUDIO HUMAN INTERACTIVE PROOF BASED ON TEXT-TO-SPEECH AND SEMANTICS

摘要

The text-to-speech audio HIP technology described in some embodiments herein uses different correlated or uncorrelated words or sentences generated through a text-to-speech engine as audio HIP tasks. The present technology can apply different effects in a text-to-speech speech synthesizer that utters sentences to be used as HIP task strings. Different effects include, inter alia, spectral frequency warping; Vowel duration warping; Add background; Echo addition; And variations in duration between words. In some embodiments, the technique modifies a set of parameters to prevent an automated speech recognition tool from training the model using previously used audio HIP tasks, once the model is trained, It can be used to recognize future audio HIP challenges generated by. Additionally, in some embodiments, the present technology introduces the requirement of semantic understanding in HIP tasks.