Overview

Unaccompanied vocal music is a central part

of Western art music, yet it requires excellent

skills for singers to achieve proper intonation.

In this contribution, we analyze intonation

deficiencies by introducing an intonation cost

measure that can be computed from choir re-

cordings and may help to assess the singers’

intonation quality. With our approach, we

measure the deviation between the recor-

ding’s local salient frequency content and an

adaptive reference grid based on the equal-

tempered scale. The adaptivity introduces in-

variance of the local intonation measure to

global intonation drifts. In our experiments,

we compute this measure for several recor-

dings of Anton Bruckner’s choir piece Locus

Iste. We demonstrate the robustness of the

proposed measure by comparing scenarios

of different complexity regarding the availa-

bility of aligned scores and multi-track recor-

dings, as well as the number of singers per

part. Even without using score information,

our cost measure shows interesting trends,

thus indicating the potential of our method for

real-world applications.

Towards Measuring Intonation Quality of Choir Recordings:A Case Study on Bruckner’s Locus Iste

Christof Weiß, Sebastian J. Schlecht, Sebastian Rosenzweig, and Meinard Müller

Intonation

scenarios:

Intonation cost measure

based on 12-tone

equal-tempered scale [2]:

with optimal, cost-minimizing grid shift and Gaussian deviation =16 cents

adaptive grid, shift

Music scenario:

Anton Bruckner,

Gradual Locus iste

Multi-track recording [1]

16 individual singers

4 sopranos S1… S4

4 altos A1… A4

4 tenors T1…T4

4 basses B1… B4

4-p

art

po

lyp

ho

ny

Fixed grid:

good intonation

Fixed grid:

global offset

Fixed grid:

global drift

Adaptive grid:

global drift + local deviations

Adaptive grid:

global drift

An Intonation Cost Measure

References & Acknowledgments

This work was supported by the German Research Foundation (DFG MU 2686/12-1). The Internatio-

nal Audio Laboratories Erlangen are a joint institution of the Friedrich-Alexander-Universität Erlan-

gen-Nürnberg (FAU) and Fraunhofer Institut für Integrierte Schaltungen IIS. We thank Helena Cuesta

and colleagues from UPF Barcelona for creating and publishing the Choral Singing Dataset.

[1] H. Cuesta, E. Gómez, A. Martorell, F. Loáiciga: “Analysis of intonation in unison choir singing.”

In Proceedings of the International Conference of Music Perception and Cognition (ICMPC), 2018.

[2] T. Nakano, M. Goto, Y. Hiraga: “An automatic singing skill evaluation method for unknown

melodies using pitch interval accuracy and vibrato features.” In Proceedings of the Annual

Conference of the International Speech Communication Association (INTERSPEECH), 2006.

[3] J. Salamon, E. Gómez: “Melody extraction from polyphonic music signals using pitch contour

characteristics.” IEEE Transactions on Audio, Speech, and Language Processing, 2012.

Selected ScenariosLocal Intonation Quality: Choral Singing Dataset [1] Global Intonation Quality: Different Recordings

Commercial

Recordings

Choral

Singing

Dataset [1]

Synthetic

Examples

fine

note

orig

orig

Estimating frequencies 𝑓 and amplitudes 𝑎 from salience representation [3]

Different conditions / technical scenarios (A … D):

Score constraints for frequency estimation or no score constraints

Individual tracks from multi-track recording [1] or mixed signal

Only first voices of each part (S1, A1, T1, B1) or all voices

Harmonic tones*, detuned 𝑑 = 0 cents

Harmonic tones*, detuned 𝑑 = 15 cents

Harmonic tones*, detuned 𝑑 = 30 cents

Choir Samples SibeliusSounds

(C) First voices 𝑥SATB1, corrected 𝑥

(C) First voices 𝑥SATB1, corrected 𝑥

(C) First voices 𝑥SATB1, original 𝑥

(D) All voices 𝑥SATB, original 𝑥

All voices 𝑥SATB, with strong reverb

Internet Archive 2013

Philharmonia Vocalensemble 1979

Chor des Bayerischen Rundfunks 2012

Choir of St John's College 1996

NDR Chor Hamburg 2000

*16 partials with decaying amplitudes

Piano roll with deviations of first voices S1, A1, T1, B1

(A) First voices S1, A1, T1, B1 (individual tracks), score constraints

(B) First voices S1, A1, T1, B1 (individual tracks), no score constraints

(C) First voices SATB1 (mixed signal), score constraints

(D) All voices SATB (mixed signal), score constraints