An efficient phonation-driven control system using laryngeal bioimpedance and machine learning

Donati, Eugenio ORCID: (2022) An efficient phonation-driven control system using laryngeal bioimpedance and machine learning. Doctoral thesis, University of West London.

[thumbnail of Eugenio_Donati_PhD_Thesis(post_viva).pdf]
Eugenio_Donati_PhD_Thesis(post_viva).pdf - Accepted Version

Download (15MB) | Preview


The extraction and conversion of human voice information are crucial in several applications across multiple subject areas such as medicine, music technology and human-computer interaction. The presented research employs the variation of laryngeal bioimpedance, measured during phonation, for extracting and processing voice information. Compared to sound recordings and microphones, bioimpedance readings deliver a much simpler signal, allowing fast and computationally non-taxing processing. In the first stage of this research, a novel system for measuring laryngeal bioimpedance was designed and built. The circuit design was implemented with a multiplexed sensor system based on multiple electrode pairs to allow self-calibration of the sensors and increase usability and applicability. In the following stage, the resulting device was used to generate a novel dataset of laryngeal bioimpedance measurements for the distinction of speech and singing. This was then used in the training and deployment of an Artificial Neural Network using the Mel Frequency Cepstrum Coefficients of the recorded bioimpedance measurements. A real-time system for converting voice into digital control messages was developed and presented as the third stage of this research. The system was implemented using the MIDI protocol for using voice to control hardware and software electronic instruments. The thesis then concludes with the integration of the complete system. The conducted research results in a self-calibrating device for the measurement of laryngeal bioimpedance which delivers an fast and efficacious real-time voice-to-MIDI conversion. In addition, the creation of a unique dataset for the distinction of singing and speech allowed the deployment of real-time classification system. Collectively, the proposed system improves applicability and usability of laryngeal bioimpedance and expands the existing knowledge in the distinction of speech and singing.

Item Type: Thesis (Doctoral)
Identifier: 10.36828/spja5598
Subjects: Computing
Music > Music/audio technology
Depositing User: Eugenio Donati
Date Deposited: 15 Jun 2023 12:57
Last Modified: 15 Jun 2023 13:48


Downloads per month over past year

Actions (login required)

View Item View Item