I was very pleased to attend the Voice Measurement & Applications workshop last week, held in London in the Speech, Hearing and Phonetic Sciences department at UCL. The workshop revealed how recent advances in audio, signal processing and machine learning are producing new applications of voice measurement. Many interesting projects were presented, and existing challenges in these fields were discussed.
Although it is not straightforward to design appropriate conversation training tasks, I felt proud that our Overlap project is working with real conversational material rather than isolated units of speech. Extended vowel phonations have commonly been used in clinical applications (for instance, asking someone diagnosed with Parkinsons to say ‘aaaaaah’). The simplification of diagnostic speech material was criticised during the workshop, however, and further to this, Dr Elina Tripoliti showed that it sometimes led to measures of voice quality which contradicted measures of real connected speech.
The workshop primarily addressed the topic of accurately measuring acoustic characteristics of the voice, assessing its quality, and tracking its changes over time. Our Overlap project is rather wider in scope, however. We do address topics of voice production, nonetheless we are more directly concerned with the communicative difficulties that cochlear-implant users face in day-to-day situations.