What went wrong with Melodiya's CD of Richter playing Schubert D894?

Nobody seems to have noticed - not even the recording editor at Melodiya - but something went badly wrong with a recent CD of Richter playing Schubert's Sonata D894. In this post I'm going to try and establish what exactly happened, and how to recover the performance from the mangled state it was issued in by Melodiya.

The CD I'm talking about is CD 2 in a 4 CD set called "Richter plays Schubert Live Released for the first time", an early part of the celebrations of the 100th anniversary of Richter's birth. The CD number is MEL CD 10 02231/2 and the recording details are correctly given as May 2, 1978 in the Moscow Conservatory.

The problem is, the last movement of the Schubert Sonata D894 lasts for 13m 20s on this CD, including clapping, but all other recorded performances of this sonata by Richter are considerably shorter. This table shows the length of each performance, with clapping removed -

Place	Date	Timing
Snape Maltings	27/09/1977	07m 38s
Snape Maltings (encore)	27/09/1977	07m 42s
Moscow Conservatory	02/05/1978	12m 50s
Moscow Conservatory	03/05/1978	07m 20s
Genova	26/01/1989	08m 08s
Bologna	30/01/1989	08m 00s
Angers	09/03/1989	07m 58s
London	20/03/1989	08m 03s
Chichester	27/03/1989	07m 45s

Even allowing for some of these recordings coming from cassettes, it is clear that the Moscow Conservatory performance of 02/05/1978 is unique in that it lasts so much longer than all the other performances.

It is certainly not the case that the other recordings have had a repeat omitted, because most of them are unedited audience recordings.

The only explanations that make sense are -

• Richter for some reason, on this occasion only, played part of the movement twice
• The recording has been edited

As it happens, I have an audience recording of the concert from 02/05/1978 in addition to the Melodiya CD. The audience recording is clearly playing flat, and runs for just over 8m 10s. If nothing else, it proves that Richter did not play for 5 minutes longer than usual, as the Melodiya CD suggests. Therefore the only explanation is that the Melodiya editor has made a mistake.

Given that this is supposed to be a live recording, there is no real justification for editing. It is not clear why the editor would choose to do anything other than simply present the recording unedited.

Bearing in mind that the recording was made just before the age of digital recordings, it is very likely that the concert was recorded onto magnetic tape. The Brilliant recording of the same Schubert Sonata on the following day lasts for 45m 43s, and it is likely that a few seconds between movements have been silently removed.

So it is just possible that the recording was made on a tape that lasted 45 minutes, or probably 45 minutes and a few seconds extra. My experience of reel to reel tapes was that an 1800 foot tape, nominally 45 minutes long, would be likely to last about 45 m 40 seconds. Engineers are likely to start recording well before the pianist sits down at the piano, for the sake of safety. So it is possible that the concert was recorded across 2 tapes, and a join was considered necessary somewhere in the last movement. If this is the case (and it is a big if) it is surprising that the engineer did not simply take the whole of the last movement from the second tape. We will never know.

Now let's see if we can find which part of the finale has been duplicated. Here is the whole movement from the Melodiya CD loaded into the Nero Wave Editor programme -

This shows the left channel at the top, and the right channel at the bottom. What we need to find is a large chunk of duplicated material in both channels, looking from left to right.

For clarity I will concentrate on just the left channel, but everything that follows applies to the right channel as well as the left channel. Here is the left channel in the wave editor programme -

Fellow Richter fan James Lisney alerted me to the evidence of coughs as showing that there is duplication. Here is a cough from 0m 54s into the finale on the Melodiya CD. Click on the red play button to listen -



As well as the cough, you may have noticed that Richter has not completely cleared the pedal between phrases -


This exact same thing returns at 6m 31s in the finale of the Melodiya CD - the cough, and the uncleared pedal -



Here is the pedal being completely cleared from the Brilliant CD recording of the Schubert Sonata from the very next day (03/05/1978) -


So, the cough shows beyond reasonable doubt that there is duplication in this recording. But which part is duplicated? Going back to the picture of the left channel above, it is pretty clear that this part (0m 41s to 1m 21s) -

is exactly duplicated by this part (6m 17s to 6m 56s) -

So the duplication starts no later than 0m 41s into the finale of the Melodiya CD. There is also an earlier cough, or some sort of audience noise, just before 0m 10s -



This reappears at 5m 44s -



So the duplication starts no later than 9 seconds into the movement. But where does it end?

There is a sound that could be a tape join at 5m 36s -



This is repeated at 11m 11s -



If you cut out the part of the finale between the two occurrences of the possible tape join, you get a sensible result. It runs for 7m 16s compared to the 7m 20s of the Brilliant CD recorded the next day. When you put the "fixed" recording from 02/05/1978 next to the Brilliant recording from 03/05/1978 in the wave editor, you can see that the shape is pretty well the same, which suggests the notes are now correct for 02/05/1978 -

The top line is the Melodiya recording from 02/05/1978 and the bottom line is the Brilliant recording from 03/05/1978. The "fixed" Melodiya is slightly shorter and the Brilliant is recorded at a higher volume, but they appear to be showing pretty well the same sequence of notes.

Quite how the editor managed to duplicate 5m and 35s of the finale will have to remain a mystery, but you can listen to my speculatively fixed version of the finale here -

[SoundCloud wouldn't let me host my edited version of the recording because it belongs to Melodiya - fair enough! But you can find it on FireDrive instead.]

My edited version on FireDrive

Remember all I have done is remove a chunk of the original recording starting at 5m 36s and ending at 11m 11s. I would very much like to hear what people think: is this a correct fix to this problem? Please let me know by commenting on this blog or emailing me at pjotaylor@gmail.com.

Pete Taylor