Predicting Speech Intelligibility with the Regeneration of Envelope from TFS Cues for Hearing Impaired Listeners

Voice metric is an active feature of speech signal as like as biometric feature, provides several unique information of speech content and speaker identity. Speech perception is the best way of extracting relevant information from the speech input signal that makes speech understandable to the listener while also helps in identifying qualitative characteristics of the talker. However, the perception of speech is challenging sometimes due to system environment and listeners hearing loss. This study proposes a metric that can quantify the listener speech perception as a form of intelligibility. The proposed metric quantify the effects of sound presentation levels (SPL), noises as well as several sorts of distortions for different hearing loss profiles in speech perception. The human auditory system has amazing capabilities to interpret speech information using slowly varying amplitude envelope (ENV) and rapidly fluctuating temporal fine structure (TFS). Due to cochlear filtering in human hearing system, rapidly varying TFS contained speech sometimes leads to envelope recovery which is known as recovered envelope. To investigate the relative contribution of recovered envelope, the proposed metric evaluates the speech intelligibility using recovered ENV cues under diverse environmental conditions. In the end to verify the proposed metric, the predicted scores are compared with the behavioral scores of the corresponding studies. Implications of this work can be assessed in designing cochlear implants and hearing aids.