C4DM Seminar: Polina Proutskova - VocalNotes: Investigating the Perception of Note Pitch and Boundaries through Varying Transcriptions of Vocal Performances from Five Musical Cultures
QMUL, School of Electronic Engineering and Computer Science
Centre for Digital Music Seminar Series
Seminar by:
Polina Proutskova, BBC/QMUL, UK, proutskova@googlemail.com
Date/time: 15th of March, 12:00-13:00
Location: GC103, first floor of Graduation Centre, Mile End Campus
Title: VocalNotes: Investigating the Perception of Note Pitch and Boundaries through Varying Transcriptions of Vocal Performances from Five Musical Cultures
Abstract: The VocalNotes project investigated how expert traditional music listeners conceive of notes in vocal performances by studying similarities and differences in their transcriptions. Teams of experts from five musical traditions (Japanese folk song, Chinese bangzi opera, Russian traditional village singing, Alpine yodelling, and Romaniote Jewish chanting) each transcribed ~10 minutes of vocal recordings from their culture, where manual transcription consisted of segmentation and note pitch correction, starting from an automatically extracted pitch curve. The experts then compared their independent transcriptions and looked for factors which could have led to disagreements.
Western staff notation is not suitable for investigating such variances, because it does not represent sufficiently fine gradations of pitch and timing. We therefore used tools that allowed more precise annotations, namely Tony for segmentation, and Sonic Visualiser for note pitch correction and transcription comparison.
We found that overall agreement was prevalent and the concept of note was generally applicable for analysis of vocal performances. Yet in some contexts disagreements were abundant, with the note concept reaching its limits. We identified four primary contexts which led to disagreements across several musical cultures: 1) differences in cultural knowledge between the transcribers, 2) differences in interpreting syllabic boundaries, 3) intra-syllabic pitch changes, and 4) “voice splash” - abrupt pitch changes caused by vocal techniques or used as an expressive device.
The VocalNotes dataset, containing the audio of the musical fragments, annotations, and song documentation, has been published for replicability and further research.
I presented to the Music Cognition Lab at QMUL the prospective methodology and the outcomes from a pilot study at the outset of the VocalNotes project, and we had a very insightful discussion. This time I will report on our findings (with examples), outline potential follow-up studies and would appreciate the group’s opinions and suggestions.
Bio: Polina Proutskova works as a senior research engineer / data scientist at the BBC R&D, conducting research on computational approaches to singing voice in her spare time. She was previously a postdoc at c4dm and did her PhD at Goldsmiths, which investigated cross-cultural vocal production through machine learning and an ontological knowledge elicitation study. Polina created singing voice datasets and founded an international academic forum on Singing Voice & AI. She is an actively performing singer and a vocal ensemble leader, with experience in ethnomusicological field research and in vocal pedagogy.