A Manual Bar-by-Bar Tempo Measurement Protocol for Polyphonic Chamber Music Recordings: Design, Validation, and Application to Beethoven's Piano and Cello Sonatas

The Thesis

Standard beat-detection software — tools designed for monophonic audio or clean studio recordings — breaks down systematically on historical polyphonic chamber music, where multiple instruments, expressive timing, and poor recording quality confuse automated algorithms. This paper's contribution is methodological: a cumulative lap-timer protocol that records bar-level timestamps with millisecond precision, preventing the error drift that plagues simpler manual approaches. The result is a public dataset covering over one hundred recordings of Beethoven's five piano-cello sonatas spanning 1930–2012. The real value is infrastructural: researchers in music cognition, performance studies, and computational musicology now have a validated, high-resolution tempo corpus they could not get any other way. The catch is that the method is labor-intensive by design, which limits how far it can scale.

Catalyst

Open, high-quality performance datasets for classical music have become a bottleneck for empirical musicology, and recent work in music information retrieval has exposed the limits of automated tools on pre-war recordings. The public availability of streaming archives and digitized historical recordings has made large-scale study possible in principle — but only if the underlying tempo data can be reliably extracted. This paper addresses that gap directly, and the release of both the dataset and analysis code makes it immediately usable by other researchers.

What's New

Earlier approaches to tempo extraction relied on automated beat-detection systems such as those built into libraries like librosa or Sonic Annotator, which use onset-detection and periodicity analysis tuned for modern, often monophonic audio. Those systems suppress or misread expressive timing phenomena — rubato (deliberate tempo fluctuation for emotional effect), fermatas (held notes), and gradual accelerations or decelerations — that are central to classical performance practice. This paper replaces automated detection with a cumulative timestamp architecture, where each measurement builds on the previous one in a way that contains and corrects its own errors, and validates the output against internal consistency checks.

The Counter

This paper documents a careful solution to a narrow problem: getting reliable tempo data out of old chamber music recordings. That is a real problem, but the audience is small — empirical musicologists working specifically on Western classical performance, a field that has never commanded large research budgets or broad readership. The methodological contribution, a lap-timer with cumulative timestamps, is genuinely sensible, but it is also not surprising; metronomists and music analysts have been timing recordings by hand for decades, and the novelty here is mostly the formalization and the public data release. The dataset covers five sonatas and roughly a century of recordings, which is meaningful for those five sonatas and essentially nothing else without additional annotation labor. More importantly, the paper does not validate its tempo measurements against any ground truth — there is no synthetic test case where the true tempo is known, which means the claimed millisecond precision is an input precision, not an output accuracy. Finally, calling manual annotation 'a principled response' rather than a workaround is a reframing that should not go unchallenged: the long-term solution is better automated tools, and this dataset, while useful, does not itself advance that goal.

The Paper