Assessing the relationship between short-term nasalance measures and phonetic transcription

A common question in speech science is how well instrumental measures correspond to listener perception. Here, we evaluate the relationship between nasalance and listeners' transcription of consonants. Nasalance is the ratio between the energy of two microphones separated by a divider at the upper lip. Most studies employing nasalance have obtained measures over long units of speech such as sentences or paragraphs. However, the system used here, the Kay Pentax Nasometer, outputs a nasalance measure every 8 ms, permitting a more fine-grained assessment. We drew on a dataset of Greek-learning children with normal hearing (NH) and cochlear implants (CI) producing words with initial nasals; voiceless stops; and voiced stops, which may be prenasalized in some speakers of Greek. Initial observations revealed cases where the transcription did not correspond as expected to the nasalance signal; for example, a sound transcribed as nasal might show zero nasalance in the closure. Our presentation explores these cases, taking into account the amplitude envelope in two microphone signals along with the nasometer signal. We observe that zero values of nasalance do not necessarily reflect an absence of nasal energy, but nasal energy that is not sufficiently strong or greater than the oral energy.