CC BY-NC-ND 4.0 · Laryngorhinootologie 2019; 98(S 02): S305
DOI: 10.1055/s-0039-1686328
Poster
Otology
Georg Thieme Verlag KG Stuttgart · New York

Speech differences between CI users with pre- and postlingual onset of deafness detected by speech processing methods on voiceless to voice transitions

T Arias Vergara
1  Ludwig-Maximilians-Universität München, München
,
S Gollwitzer
1  Ludwig-Maximilians-Universität München, München
,
JR Orozco-Arroyave
2  University of Antioquia, Medellin, Colombia
,
JC Vasquez-Correa
3  Friedrich-Alexander University Erlangen-Nürnberg, Erlangen
,
E Nöth
3  Friedrich-Alexander University Erlangen-Nürnberg, Erlangen
,
C Högerle
1  Ludwig-Maximilians-Universität München, München
,
M Schuster
1  Ludwig-Maximilians-Universität München, München
› Author Affiliations
The authors acknowledge to the Training Network on Automatic Processing of Pathological Speech (TAPAS) funded by the Horizon 2020 programme of the European Commission.
Further Information

Publication History

Publication Date:
23 April 2019 (online)

  

Introduction:

The onset of deafness affects speech in different ways. Speech differences of Cochlear Implant (CI) users with pre- and postlingual deafness are examined using acoustic features extracted automatically from speech.

Methods:

Utterances of 22 prelingual (15 up to 71 years old) and 22 postlingual CI users (15 up to 78 years old) were analyzed. All patients read 97 words, which contain every phoneme of the German language in different positions within the words. Speech analysis is performed in the transitions from voiceless to voiced sounds that mark the precise control of speech movement patterns. To extract the transitions, we search for the boundary between voiceless and voiced sounds using the fundamental frequency with a constant segment of 80 ms to the left and right. The feature set includes 13 Mel-Frequency Cepstral Coefficients and their 1st and 2nd derivatives. The mean, standard deviation, skewness and kurtosis are computed from the descriptors, forming a 156-dimensional feature vector. Wilcoxon signed-rank test is used to find differences between the pre- and postlingual groups.

Results:

The Wilcoxon signed-rank test was performed for each descriptor and significant differences between the pre- and postlingual groups (α < 0.05) were found in 8 of the 156 features. Additionally, a support vector regressor was trained to evaluate the age independence of the selected features. According to the results, there was not a strong correlation between the age of the speakers and the selected features (ρ < 0.40).

Conclusions:

Speech patterns differ significantly between pre- and postlingual CI users at the transitions of voiceless to voiced sounds. Other acoustic features are to be examined and considered in the rehabilitation after cochlear implantation.