Vowel intelligibility analysis in singing using voice directivity features (vor Ort)
* Presenting author
Abstract:
Intelligible pronunciation is a key aspect in classical singing and a necessity for correct text delivery. On the contrary, the modification of vowels in classical singing is used to achieve better resonance and ease at high pitches. This can result in diminished vowel intelligibility, which has been reported in literature especially for sopranos and tenors. Still, it seems not possible to clearly identify up to which pitch vowel intelligibility is most likely feasible and there exist contradictory results. Common approaches for vowel identification use linear prediction coefficients or mel-frequency cepstral coefficients. For both, the accuracy is reported to bepitch-dependent and decreases with increasing pitch. We present a novel approach to vowel identification by using voice directivity features. The voice directivity is affected by the size of the mouth opening. Preliminary results show that a distinction between back vowels and front vowels is possible based on features of voice directivity.