musical archaeology
by
| JonathanBerger CCRMA, Stanford University |
An analysis and transcription of the 1889 cylinder recording of Johannes
Brahms piano performance of a segment from his First Hungarian Dance.
Please note:
This site is currently undergoing a long overue upgrade. Stay tuned
the new site with all sound examples will be up on January 1 2001
Happy New Year.
For the first time ever this historic musical document has been accurately deciphered. The following account includes aspects of musicology, mathematics and detective work.
The work involved developing a denoising method, painstaking removalof layers of the recording in order to reveal music 'hidden' within thenoise, and reconstruction of the analyzed and transcribed layers.
Transcription was followed by careful analysis of the numerous performance
nuances, agogic inflections, improvised segments and added elaborations.This
data was used to reconstruct the entire, original four hand versionof the
piece for Disklavier.
Although the recording technology lacked the sensitivity to derivedynamics
or pedal markings, the timings suggested distinct performancetrends that
used to create what I believe, is a pretty good approximationof how Brahms
would have played the piece. As sacreligious as it may sound,the result is
not only a representation of nineteenth century performancepractice, but
a computer generated performance of how Brahms would haveplayed with himself!
Here are examples of the process, from the original signal, tothe denoised version, and finally a segment with an overlay of our transcription (.wav format sound files):
This is the same phrase with a synthetic sound overlaid upon thevoice that was 'excavated' by the denoising process.
The Mystery of the Brahms-Edison Cylinder
The Brahms cylinder recording has been a source of excitement and great frustration for many over the past decades. Excitement in that it is aunique document both in terms of the technology and, of course, in termsof the rare primary source material for Brahms research. Unfortunatelythe recording has been extremely frustrating to musicologists because ofthe unbearable amount of noise that literally masks the music. In additionto the enormous amount of noise in the recording, tracing with certaintythe history of the Brahms cylinder has to date, left numerous questionsunanswered. The combination of noise and questions of documentation regardingthe recording has been a source of raging controversy.
Our work on denoising the recording has answered many of these questions.
Despite other attempts at filtering and enhancing the recording, thepoor quality of the cylinder recording resulted in a general concensusthat the recording was not of significant musicological or historical value.In Gregor Benko's words:
" any musical value heard [in the cylinder recording as released byon LP by the International Piano Archives] can be charitably describedas the product of a pathological imagination."
By applying orthoginal trigonometric and wavelet-based analysis techniques,
we were able to reconstruct enough meaningful musical data for us to challenge
this long held view.
Who is speaking?
Numerous writers, scholars and amateurs alike, attribute the voice tobe that of Brahms introducing himself with the words
"I am Doctor Brahms, Johannes Brahms."
However there are a number of factors that cast doubt upon this asumption.
However the only mention of the recording (in published memoires ofFellinger's son) states that Brahms was introduced.
Brahms indeed received an honorary doctoral degree late in his life,but rarely, if ever, referred to himself as a doctor.
Consider the time between the announcement and when the playing commences. It seems rather improbable that the same person would segue so immediately into the performance. The limitations of the recording technology makesthis all the more unlikely
Finally, the distorted sentence that precedes the 'Dr, Brahms' sentence has been transcribed as being in English by some (as " Dr. Edison"), andas German by others ("By Herrn (?) Dr. Fellinger (?)"). I believe thatthe introduction is the voice of Wangemann, and the actual transcript is:
[garbled] ..."? ...house of Herr Doctor Fellinger, I have Dr. Brahms,Johannes
Brahms."
Why did Brahms record the First Hungarian Dance?
The Hungarian Dances were enormously popular during Brahms' lifetime.Whether or not this had bearing on Brahms' choice of the first dance forthe recording is open to conjecture. Perhaps the current popularity ofthe pieces hint at the extent to which they were heard before recordingtechnology. There are versions for orchestra, piano four hand, two pianos,violin and piano, various chamber ensembles, marching band, and more. Thereis even a late 19th century account of an all girls performance of an eightpiano arrangement of the piece.
Brahms himself arranged the piece from the original piano 4-hand version to its more popular orchestra version, and to the solo piano version that he chose to record. This contrasts his harshly critical conservatism regarding arrangements that is noted in a letter to Henschel in 1881,in which Brahms wrote "I prefer to retain my ears and know what is a pianoforte piece and what is an orchestral piece...."
From George Henschel, "Personal Recollections of Johannes Brahms", Boston
1907, pp. 80-81.
Features of Hungarian music formed an integral part of Brahms' musical
style. Brahms performed Hungarian gypsy music on concert tours with theviolinist
RemÝnyi earlier in his career. Originally composed between1852 and
1864 the four-hand set of twenty dances was transcribed by Brahmsfor two
hands in 1872 and for orchestra in 1874. A violin and piano arrangementwas
prepared by Joachim. In addition to the the Hunagarian dances Brahmswrote
the Variations on a Hungarian Song (op. 21 no. 1) in 1857 and the11 Zigeunerlieder
op. 103 in 1887. The Piano Quartet, op. 25 (1861) featuresa Rondo alla zingarese
and a set of gypsy poems in translation that comprisefour of the op. 112
quartets (1891).
The denoising algorithms were developed by Prof. Ronald Coifman,Prof. Maxim
Goldberg, Dr. Igor Popovic and myself. Initial research wasdone at Yale University,
and is currently continuing at
FMAH, inc.
For a complete description of the denoising method see our papers inthe Journal of the Audio Engineering SocietyAES Paper,and the Proceedings of the International Computer MusicConference, 1995 and 1996.
In 1946 Denis Gabor, the inventor of the hologram, proposed that conventional representations of signals as continuous functions of time could be supplanted by a representation that combined time and frequency. Considering the finite information present in actual signals he divided this information planeinto 'cells'. Each signal can be represented in many ways. It is this basicidea, and the resulting flexibility of being able to represent sound inboth the time and frequency domains, that is at the core of our method.
An other denoising method was attempted on the Brahms cylinder was developed at the Vienna Phonogramarchiv byWerner Deutsch.
In addition to Brahms' liberal rubato and some protracted fermatti,the performance
lapses into improvisation at a number of points.
Brahms tempo for the piece is mm 80 = eighth note, considerably slower than any recent recording.
The reconstructed sound file allowed us to precisely measure and compare various temporal aspects of the performance. While the overall durationsof the three 'dotted rhythm' phrase groups are close in duration (mm 13-24is 10.82 seconds, mm 25-36 is 9.28", and mm 37-48 is 9.89"), the internallengths and proportions of each measure unit is remarkably flexible. Thelengths of measure units range from .69" (m 21) to .93" (m. 33).
Brahms tends towards consistent 'underdotting'. In fact, the eighthnote exceeds the doration of its preceding dotted quarter in measures 14and 19, and approaches equal durations to the dotted-quarter in mm 20,34, and 40. Brahms gives the dotted quarter its full value only once inthe entire piece, in measure 38.
This surprising feature of Brahms' performance runs contrary to thepersistent tendency to extend the longer duration at the expense of theshorter rhythmic values. This tendency (known as over-dotting when doneconsciously) was documented in numerous treatises on performance sincethe 17th century.
sixteenth note patterns.
The middle section of the recorded segmnent is comprised largely ofthree sixteenth note elaborative figures characterized by an upper neighborframed by chord tones. We were able to detect the structural descendingline however, the analysis method did not allow us to resolve the frequencieswith the fine granularity necesary to measure each note of the figures.A sumary of our analysis of this section reveals that Brahms was far moreaccurate placing first and second beats in measures 49-71 than he was placingthe eighth notes following the dotted-quarters.
With the above performance data I proceeded to reconstruct the entireoriginal
four hand version for the Yamaha Disklavier. This work was commissionedby
the Milwaulkee Symphony and premiered at their Brahms Centenary Festivaland
at the TED IX conference.
The Transcription and reconstruction
This .wav file of the midi reconstruction thatyou are hearing take all the data that was taken from the analyses of thecylinder and map them into the entire piece. Although the recording technologylacked the sensitivity to derive dynamics or pedal markings, the timingssuggested distinct performance trends that used to create what I believe,is a pretty good approximation of how Brahms would have played the piece.In fact, some of the performances you hear use the data mapped onto the4-hand version, so you can even imagine Brahms playing with himself!
If you have a disklavier (or good sampled piano synth) try this
.midifile
1877 was a revolutionary year in the history of audio technology.Edison invented
the cylinder 'phonograph' and Emile Berliner invented themicrophone (which
Bell Telephone acquired). Ten years later Berliner inventedthe'gramophone'
which was played a flat media through an acoustic amplificationhorn.
In 1889, the year that Brahms recorded the Hungarian Dance, magneticwire recording was invented (by Valdemar Poulson in Denmark).
By 1899 recording technology was considered marketable.
However much was lacking, most importantly, a method to commerciallyreproduce the cylinders. Berliner's flat disk system made it easier todevelop duplication and distribution methods.
Early recording technologies were problematic in their variablilityin speed. This made us consider the effect of the warp on the cylinder,and the variance of the drive mechanism in the cylinder spindle in doingthe transcription. The cylinders were self destructive - each play wouldshave down more and more of the signal.
Despite the limitations in audio quality and durability, it wasn't long before Edison developed a publicity and advertising campaign in which hehad an opera singer and a phonograph behind a screen and challenged theaudience to identify which was real and which was not. (Recall that audenceswere reported to react with fright at the 'realism' of black and whitesilent films projected from unstable projectors).
For further information on Wavelets WaveletIndex
For further information on CCRMA
TheCenter for Computer Research in Music and Acoustics.
© 1999 Jonathan Berger, CCRMA, Stanford University.
All Rights Reserved.
Portions of the software described are patented byFMAH
inc., Hamden, CT This page was created and is maintained by JonathanBerger