Designing a Pneumatic Bionic Voice Prosthesis - A Statistical Approach for Source Excitation Generation

Farzaneh Ahmadi, Tomoki Toda


This study follows up on our pioneering work in designing a Pneumatic Bionic Voice (PBV) prosthesis for larynx amputees. PBV prostheses are electronic adaptations of the traditional Pneumatic Artificial Larynx (PAL) device. The PAL is a non-invasive mechanical voice source, driven exclusively by respiration and with an exceptionally high voice quality. Following the PAL design closely, the PBV prosthesis is anticipated to substitute the medical gold standard of voice prostheses by generating a similar voice quality while remaining non-invasive and non-surgical. This paper describes a statistical approach to estimate the excitation waveform of the PBV source using the PAL as a reference. A Gaussian mixture model of the joint probability density of respiration and PAL voice features is implemented to estimate the excitation waveform of the PBV. The evaluation on a database of more than two hours of continuous speech shows a close match between f0 pattern and mel-cepstra of the estimated PBV source and the PAL. When used to re-synthesize the original speech, the intelligibility of the PBV speech remains high and is scored 7.1±0.4 compared to 7.9±0.15 of the original PAL source.


 DOI: 10.21437/Interspeech.2018-1043

Cite as: Ahmadi, F., Toda, T. (2018) Designing a Pneumatic Bionic Voice Prosthesis - A Statistical Approach for Source Excitation Generation. Proc. Interspeech 2018, 3142-3146, DOI: 10.21437/Interspeech.2018-1043.


@inproceedings{Ahmadi2018,
  author={Farzaneh Ahmadi and Tomoki Toda},
  title={Designing a Pneumatic Bionic Voice Prosthesis - A Statistical Approach for Source Excitation Generation},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={3142--3146},
  doi={10.21437/Interspeech.2018-1043},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1043}
}