"Polyphonic Singing Datasets for MIR Research" » History » Version 2

Polina Proutskova, 2020-10-06 04:00 PM

1 1 Polina Proutskova
h1. "Polyphonic Singing Datasets for MIR Research"
2 1 Polina Proutskova
3 2 Polina Proutskova
Helene Cuesta and Sebastian Rosenzweig
4 2 Polina Proutskova
5 1 Polina Proutskova
h2. Abstract:
6 1 Polina Proutskova
7 1 Polina Proutskova
We showcase two recently published datasets of polyphonic a 
8 1 Polina Proutskova
cappella vocal music. Both publicly available datasets form exciting 
9 1 Polina Proutskova
research scenarios for music information retrieval (MIR) tasks such as 
10 1 Polina Proutskova
(multiple) F0-estimation or source separation in vocal music. First, we 
11 1 Polina Proutskova
present the Erkomaishvili dataset, a curated corpus of traditional 
12 1 Polina Proutskova
Georgian vocal music. The corpus is based on historic tape recordings of 
13 1 Polina Proutskova
three-voice Georgian songs performed by the former master chanter Artem 
14 1 Polina Proutskova
Erkomaishvili. Second, we present Dagstuhl ChoirSet, a multitrack 
15 1 Polina Proutskova
dataset of "Western" a cappella choral music. The dataset includes 
16 1 Polina Proutskova
recordings of an amateur vocal ensemble performing two choir pieces in 
17 1 Polina Proutskova
full choir and quartet settings. The audio data was recorded during an 
18 1 Polina Proutskova
MIR seminar at Schloss Dagstuhl using different close-up microphones to 
19 1 Polina Proutskova
capture the individual singers' voices.