Authors: Marieke M.W. ten Hoor1 , Floris Rotteveel 2, Bert Maat, Deniz Başkent 2 , Etienne Gaudrain 3
1 University of Groningen, 2 University Medical Center Groningen, 3 Lyon Neuroscience Research Center
Background: In addition to speech intelligibility, voice recognition is an essential part of human communication. Therefore, with a cochlear implant (CI), individuals must not only be able to distinguish different speech sounds, but also to differentiate vocal characteristics of the talkers, such as voice pitch (F0) or speaker size, which is reflected in speaker’s vocal-tract length (VTL). Whereas normal hearing (NH) individuals can make fine judgements about relative speaker sizes by listening to voice cues, this ability is hindered in CI users. However, according to theoretical CI sound processing, VTL differences are expected to greatly affect the output signal, yielding features that should be perceived by CI listeners to some degree. This leads to the question whether the implant’s output is conform to the theory, and whether the representation of VTL differences is as salient as expected. Therefore, the current study aims to characterize the representation of VTL differences in the CI output signal.
Methods: Systematic variations of VTL (-12 to +12 semitones) will be created in Dutch syllables (consonant–vowel), words, and short sentences using the STRAIGHT program in MATLAB. An automated data acquisition system will be set up to directly record the output of an implant-in-a-box device (16 channels) from Advanced Bionics while presenting the manipulated speech stimuli. The representation of VTL differences in the CI output signal will be characterized by means of feature extraction.
Results and Discussion: Spectral features are expected to be most important as VTL is associated with formant frequencies in the spectral envelope of a speech signal. Eventually, the salience of the extracted features can be used to predict the behavioral outcomes previously obtained in implant users.