PORTEFEUILLE

People, who have undergone laryngectomy as a treatment of the laryngeal cancer, loose the ability to speak. The vocal function can be recovered partially through medical rehabilitation, however, with a loss of speech intensity and a poor quality of the pitch (fundamental frequency of speech). To overcome these deficiencies a vocal aid demonstrator for laryngectomees has been developed in the frame of the LARYNX project. Such a vocal aid improves the quality of laryngeal speech through advanced signal processing techniques and restores it in real-time with a higher intensity and quality. This will allow laryngectomees to improve their social integration.

The development of an intelligent vocal aid for laryngectomees requires multidisciplinary scientific and technical competencies and the merge of the medical expertise of voice rehabilitation and the engineering aspects of speech processing and voice reconstruction. The three research institutes CSEM, HEIG-VD and CHUV-ORL with their proven experience in the domain of biomedical engineering ensure the excellence in multidisciplinary research. The activities are situated between basic and applied research for a small business niche and the implemented approach on laryngectomees could be generalized and engender a novel line of devices for voice-impaired persons in the same way as it has occurred during the past for hearing impaired.

The project Vocal aid for laryngectomees focused on the development of a processing framework and an HW demonstration platform for the restoration of the voice quality. During the first project phase the technical feasibility has been demonstrated and preliminary results have received a very positive appreciation from speech therapist experts. Simultaneously, a market study has been performed (for more information see GRS-023/3).
The proposed approach is exclusively based on non-intrusive speech restoration by means of digital processing unit. This approach can be summarised as a decomposition of the speech signal into fundamental components, a restoration of these components and finally a re-synthesis.
The technical feasibility has highlighted two challenging problems for the successful development of such a voice restoration system. Firstly, in pathological voices the prosodic information is degraded and cannot be estimated using classical processing approaches. However, prosody is a key feature for the restoration of a natural voice and it is otherwise perceived as an artificial voice. Several approaches to preserve/restore the prosody are under development. Secondly, the estimation of the vocal tract characteristics is a difficult problem with pathological voices and classical approaches fail in such case. Specific methods, taking into account the characteristics of pathological voices and prior knowledge about speech signal have been developed and bring the development of a commercial system closer.
In parallel, several hardware configurations for the realisation of a real time demonstration system have been evaluated. A real-time system, fulfilling the expected requirements, has been realised.
Another important point is that the results obtained during the project have allowed to generate additional funding from the Swiss League against Cancer for a PhD position at CSEM. The candidate started his activities in the beginning of 2008 on the theoretical aspects of the problem. This complementary activity is an asset for the future success of the project.
Finally, the project resulted into the submission of a patent application and several publications on the topic have been edited (see below) .

"Procédé et dispositif de conversion de la voix", Rolf Vetter, Jérome Cornuz, Philippe Vuadens, Josep Sola I Caros, Philippe Renevey, European Patent No EP1710788, October 2006.
“Restoration of Natural Prosody in Pathological Voices”, F.N. Reale, J-M. Solà, Ph. Renevey and Rolf Vetter, Swiss Society of Biomedical Engineering (SSBE), annual meeting, Neuchâtel, Switzerland, 2007.

Project references on the homepages of the project partners (see links) with corresponding audio samples of patient data.
The project has been presented in the newspaper Le Temps the 25th October 2009, http://www.letemps.ch