Header

Towards Interdisciplinary, Computer-assisted Analysis of Musical Interpretation: Herbert von Karajan

Project number        P 29840 Einzelprojekt
Project lead            Peter Revers
Decision board         2017/03/06
Project on FWF website
https://pf.fwf.ac.at/en/research-in-practice/project-finder/39888

A Study of Herbert von Karajan

Project Characteristic

Analysis of Musical Interpretation: Herbert von Karajan is a three-year research project (June 2017 - June 2020) funded by the Austrian Science Fund (Fond zur Förderung der wissenschaftlichen Forschung, FWF, Project Number P 29840-G26). Headed by the Project Leader, Peter Revers (University of Music and Performing Arts Graz), it comprises an inter-disciplinary team of scholars (including several doctoral students) with a particular focus on musicology and computational perception.

Scientific Aim

The project sets out to accomplish a data-driven study in musical interpretation, connecting information from human analyses of a number of performances by Karajan, with algorithmically derived data spanning several decades of recordings. Although the project is conceived as a case study on Karajan, we wish to demonstrate the promise of data-intensive, interdisciplinary approaches to musicological studies of expressive performance, and help in establishing this as a new open standard approach in empirical musicology.

Innovative aspects

Our approach goes beyond already existing projects (e.g. the “Mazurka Project” as the first large scale project of this sort) in that we focus for the first time exclusively on orchestral music conducted by a specific performance artist. In addition to musicological and computational research results, the project has access to a unique collection of largely unpublished music and extra-musical sources (hosted by the Eliette and Herbert von Karajan Institute, Salzburg).

Methods

First of all, we focus on a comparative analysis of Karajan’s interpretations of the same composition from different time periods, following by a comparative analyses of tempi, articulation and orchestral balance in selected performances of orchestral works by Karajan and other contemporary conductors (using tools like Sonic Visualiser to pinpoint and visualize musically relevant passages and performance characteristics). Secondly, we concentrate on the aesthetic perception of a specific “Karajan profile” as conductor and media-artist. Moreover, we seek to develop some new methods for aligning performances to scores, and performances to alternative performances, to support performance (timing) analysis and comparative analysis of recordings.

Research and Results

Creation of a project homepage and a dynamic web application site

Creating a homepage for presenting our project aims, stages, staff, and results (www.karajan-research.org), as well as a second website for collecting, presenting, and sharing specific analytical data, most of which represents temporal annotation data (oma.digital). The latter also provides applications that allows external users to upload, organize and visualize publicly available data structured by composer, composition (section), and (concrete) performance. The design of the database and the source code of the web application are documented in more detail and published as an open-source project on Github (github.com/KarajanResearch/OpenMusicAnnotations).

Extensive work towards computational methods for efficient, precise, and robust temporal annotation of orchestra recordings

A) Initial investigations into using OMR (Optic Quantificational Music Recognition) for automated score linking, with largely negative results, due to the complexities of full orchestra scores.

B) Systematic studies on alignment accuracy and experimental quantification of inter-annotator agreement.

C) Improving alignment robustness by combining Dynamic Time Warping (DTW) with robust statistical methods.

Compiling and documenting a systematic Beethoven Symphony Corpus

A) manual annotation of the complete cycle of 9 symphonies by Karajan, to be used as reference for the following steps.

B) Checking and documentation of different repeat structures in this large set of recordings; manual annotation of sections, structure matching.

C) Semi-automatic transfer of annotations from Karajan reference set to other recordings via automatic audio-to-audio alignment and extensive plausibility checking.

D) Transfer to Open Music Annotations (OMA) Database.

Reconstruction of the aesthetics of music-films

Gaining a deeper understanding of Karajan’s visual aesthetics by an analysis of his music films with both quantitative and qualitative methods, starting with his early films (1964-71) and leading to later, partly more complex productions.

Analyses of selected performances

Both by a careful use of technically extracted audio features and by close listening, many aspects of musical performance have been addressed, such as timing/tempo, formal-dramaturgical shaping, timbre, dynamics, prominence of melodic lines, accentuation, legato playing, sound transformations, and considerations of orchestral-instrumental balance and disposition.

Archive research

By examining selected orchestral scores and parts that were in use during Karajan’s time in Berlin for the first time, we shed light on practical components of Karajan’s conducting that often tend to be ignored or mystified. In addition, sources from both the Karajan Institute Salzburg and the Salzburg Festival Archive, including reviews and correspondences, were and will be used to shed further light on the organizational backgrounds and the reception of Karajan’s performances.
Archiv

Doubling (“Verdopplung”) of Flute 1 from orchestral parts of Ludwig van Beethoven’s Fifth Symphony used by the Berlin Philharmonic Orchestra under Karajan, indicating in which sections the doubling instrument was supposed to play. Source: Archive of the Berlin Philharmonic Orchestra.

Interviews

We conducted interviews with several former members of the Berlin Philharmonic Orchestra in order to deepen our understanding of Karajan’s aesthetics as well as his practical approaches to working with orchestras, especially with regard to certain findings during our archive researches.

Selected Results

Scientific Methods and Insights of Computational Research

From a scientific perspective, the results produced by the computational research are in form of new scientific/computational/statistical methods, and new insights into the feasibility of automatic annotation transfer in complex orchestral recordings, as well as into the relative suitability and robustness of various audio features and feature parametrization.

Quantification of expected annotation, in(accuracy) of human time annotations (bar/beat), expected inter-annotator agreement.

In experiments with recordings of complex music of different styles made by different conductors that were mutually annotated by several human annotators, we showed, by statistical analysis, that inter-annotator variability in complex orchestral music is in a range of typically less than 50ms, which seems well suited for musicological analysis. The observed variabilities do not conform to a normal distribution, as was hitherto believed in literature (but now tested in music of this complexity). (Gadermaier / Widmer 2019)

A statistical model to obtain a quantitative measure local imprecision.

Based on the results (see above, p. 2) we proposed a new statistical method of estimating local annotation accuracy, using the concept of pooling, which gives more reliable estimates of local variance between annotators. This is also relevant with a view to transferring this information to data derived from such annotations (e.g., tempo curves).

New methods for improving alignment robustness by combining DTW with robust statistical methods.

Two robust combination algorithms based on median filtering and local spline fitting (regression) were developed and experimentally validated on orchestral music of different styles recorded by different conductors. (Grachten / Chacón / Gadermaier 2019)

Reconstruction of the Aesthetics of Music-films

Our analyses showed that Karajan’s film aesthetics was influenced by different approaches and film makers (esp. Clouzot and Niebeling), especially regarding his use of camera settings and scene changes (cuts, fading, overlays, etc.), and that the style of his movies did not change significantly after about 1973. It also became clear that the common opinion that Karajan would mainly stage himself in his movies can no longer be upheld, since the quantitative evaluation shows that in only ~15% of his movies’ screentime he is shown completely alone. This has been demonstrated in detail by the first substantial musicological analysis of the movie for Berlioz’ Symphonie fantastique. (Hebenstreit 2023)
Film Karajan dimly lit and in profile view in his music film of Hector Berlio’s Symphonie fantastique (1970/71). Taken from Hebenstreit 2023, p. 250.]

Results of performance analyses

As for the research in Salzburg and Graz, the detailed and comparative analysis with dozens of recordings per composition, some common and widespread verdicts (prejudices) concerning Karajan’ shaping of musical time and sound were given a more differentiated treatment, among others:

A) The overly simplified image of a “constant” tempo for an entire piece had to be dismissed early (considerable expressive fluctuations are sometimes present even on the level of the beat-patterns). Nevertheless, a relatively homogenous tempo shaping has been observed in many cases, particularly if compared to the recordings of many contemporary conductors, which led to an empiric reinforcement of that often-mentioned feature of Karajan’s performances. (Wozonig 2020, 2022a, 2022b, 2023a)

B) An important part of the project was the “Beethoven Symphonies Annotation Corpus”, a large set of temporal annotations relating to the complete set of 9 Beethoven Symphonies, as recorded by a number of different orchestras and conductors. One of them (1962/63) was annotated entirely manually; these annotations were than mapped to the other recordings in a semi-automatic way, relying on robust alignment algorithms. (Wozonig 2020)

C) Software Tools for finding structurally compatible performances in a large corpus of recordings, and for obtaining a visual overview of a given corpus. (Wozonig 2022b, 2023a)

D) It was demonstrated and visualized in selected examples how the parameters sound and dynamics contribute to Karajan’s shaping of certain dramaturgical developments, e.g., to create an extended, formally overarching intensification (see examples below)

E) Differentiation of the common view of Karajan as a “slow” conductor: It was demonstrated that whenever Karajan conducts relatively slow tempi with regard to performance history, this only applies to individual parts of works. Through this, Karajan often aims for a strong contrast between sections and, therefore, a specific formal-dramaturgical shaping. A specific case was highlighted by an analysis of Sibelius’s First Symphony, where it was demonstrated that Karajan’s standalone rendition of the 2nd movement was actually the result of an observance of the composer’s metronome markings. (Aringer 2019, 2020a, 2020b, Wozonig 2022b)

F) It was demonstrated that Karajan’s conception of a musical work (e.g., basic tempo choices) was often developed at an early stage, and that these conceptions would only change gradually in the course his career. Particularly revealing was the finding that if there were bigger changes in later performances, they were often connected to a stronger influence of Espressivo elements.

G) By an interdisciplinary reading of various sources, we were able to show that the orchestra and individual musicians played an important part in the performances conducted by Karajan, and that this artistic partnership was even welcomed by him in that he relied on musicians that were able to understand and share his artistic concepts. This was not only true with regard to the string section during his time in Berlin, but especially to their concertmaster Michel Schwalbé. This relativizes the popular concept of the conductor and especially Karajan as a “philharmonic autocrat”.

A A A
The three graphs illustrate the development of dynamics in “Von den Hinterweltlern” from Richard Strauss’s Also sprach Zarathustra. In contrast to the contemporary recordings by Fritz Reiner 1954 and Karl Böhm 1958, Karajan, in his 1959 recording, is much more concerned with an overarching intensification, which includes both a slightly stronger increase in dynamics up to bar 51, and a more restrained entry to the following section (esp. if compared to Reiner). X-axis: bars, y-axis: dynamics (normalized). Source: Wozonig 2022b
A Tempo graphs of four early recordings of the first movement from Jean Sibelius’ Sixth Symphony, compared with Karajan’s first recording of the piece from 1955. As the graphs illustrate, Karajan chooses a hitherto unknown approach in that he chooses a very slow tempo for the first 66 bars and a much faster (yet not extremely fast) tempo for the following section, thus shaping a distinct two-part form with a ‘slow introduction’ before the main sonata form. X-axis: bars, y-axis: beats per minute. Source: Wozonig 2023a.
A Detail of phrasing of the vocal line “Ist dies etwa der Tod” from Richard Strauss’ “Im Abendrot” (Vier letzte Lieder). Source: Revers 2022.

Imprint

Responsible for the content:

Eliette und Herbert von Karajan Institut
Imbergstrasse 6
A-5020 Salzburg

Tel. +43 (0) 662/84 11 11
Fax +43 (0) 662/84 11 11 - 14
E-Mail info@karajan.org
Web www.karajan-institut.org

CEO: Matthias Röder
CFO: Henry Ladewig


Registered trademark:
The name HERBERT VON KARAJAN/KARAJAN® is a registered trademark and trademark of the Karajan estate.

Privacy Policy

The protection of your personal data is of particular concern to us. We therefore process your data exclusively on the basis of the legal regulations (DSGVO/GDPR, TKG 2003). In this privacy policy information we inform you about the most important aspects of data processing on our website.

 

Hosting

The Amazon Web Services used by us serve the provision of the following services: Infrastructure and platform services, computing capacity, storage space and database services, e-mail delivery, security services and technical maintenance services that we use for the purpose of operating this website.

We, or our hosting provider, collect the following data on the basis of our legitimate interests within the meaning of Art. 6 para. 1 lit. f. DSGVO/GDPR data on each access to the server on which this service is located (so-called server log files). Access data includes the name of the accessed website, file, date and time of access, transferred data volume, notification of successful access, browser type and version, the user's operating system, referrer URL (the previously visited page), IP address and the requesting provider.

Log file information is stored for a maximum of 7 days for security reasons (e.g. to investigate misuse or fraud) and then deleted. Data whose further storage is required for evidentiary purposes are excluded from deletion until the respective incident has been finally clarified.

Further information about Amazon Web Services can be found on their privacy policy page: https://aws.amazon.com/privacy/?nc1=h_ls

 

Marketing measures and data collection

On the basis of our legitimate interests pursuant to Art. 6 (1) (f) DSGVO/GDPR and an order processing contract with the respective companies pursuant to Art. 28 (3) (1) DSGVO/GDPR, pseudonymous data is collected, as far as technically feasible by us, which serves us for the evaluation and optimisation of our services.

 

Your Rights

Basically, you have the rights to information, correction, deletion, restriction, data transferability, revocation and objection. If you believe that the processing of your data violates data protection law or your data protection claims have otherwise been violated in any way, you can complain to the supervisory authority. In Austria, this is the data protection authority.

© Karajan Research