Inadequate precision of "Change Tempo" effect

I am trying to restore a very old and severely damaged vynil record, which comes in two alternative copies.

I have two different takes of the same “album” (two physical discs from the same batch digitized using two diferent players). Audacity has a wonderful “remove|isolate center” effect, which can be used to separate the music (identical signal from the two records) from the damage (differing signals)

The two takes differ in tempo and I need to do a perfect peak-to-peak match of the two takes. The tempo effect is the tool to do it, but it is way too coarse to be used successfully. The record is 36 min, 26 sec long. and the digitized pcm versions differ by 4065 milliseconds in length (pregaps aligned). I cannot make audacity 2.3.3 to adjust the length precisely :imp: . The result is always either too long or too short and NEVER perfect.

The tempo effect setup dialog does not let me even to enter the correct adjustment factor, which has 4-5 digits after the decimal point or the correct target length (only two digits after decimal point allowed, while I need at least three!). It looks like the effect does not have enough numerical precision to produced the required result. Could it be fixed?

Try using the “Change Speed” effect instead.

The problem is the damage/difference is from both recordings so I don’t think there’s a simple way to subtract it out the original (or from a mix of the originals). It’s like the old problem of trying to keep the vocals/center while throwing-away the sides. It takes advanced processing to do that.

Audacity has a wonderful “remove|isolate center” effect, which can be used to separate the music (identical signal from the two records) from the damage (differing signals)

That subtracts left from right which isn’t exactly what you want to do.

However, you can invert one copy and then mix. Mixing is done by summation so inverting & summing is subtraction and that will give you the difference. I’m just not sure if you can (easily) do anything with the difference.

There might be some more advanced processing that you could do… When I’ve thought abut this, I thought about picking the quietest track moment-to-moment , or maybe pick the track moment-to-moment with the least high frequency energy.

Also, your time-alignment might be “close enough” if the speed is constant but it can’t be “digitally-perfect” because every time you sample an analog waveform it samples at a different point in the waveform. (If it turns-out to be a problem, a higher sample rate should help, but it sounds like you don’t have both copies in-hand so you can’t re-digitize.)

BTW - There is a plug-in called [u]VocAlign[/u] that might be able to time-align the two recordings, but it starts at $150 USD and it doesn’t appear to be compatible with Audacity. And, it probably doesn’t run on Linux.

This is a wrong advice. I have successfully restored some old records and I know that inverting and mixing one noisy copy with another noisy copy does not produce clean sound. This can be demonstrated by taking a mono LP record digitized in stereo and applying both “isolate center” (FAR better) and invert R + subtract L from R. The first gives you very good mono signal while the latter cancels the signal and produces even more noise.

Instead I can separate both stereo records into two L and two R mono tracks and isolate center from each L+L and R+R pair separately. This gives better sounding L and R which can be combined back to stereo and processed further. E.g. you can invert-mix the resulting sterile sound with one of the copies to get all noise plus some residue music and use maximum force declicking/noise removal without fear to damage the main part of the music signal. Summing the results with the “isolated-centers” version produces a more natural vynil sound without clicks and scratches (If everything goes well). Invert mixing that with the original can produce a file with only clicks, which can be reviewed/filtered before invert-mixing ONLY the clicks with the original, which gives a perfect restoration copy with all voice nuances.

My problem is time-aligning two source PCM streams from two physical LPs containing the same record but having different sets of damages. They were made with players that have slightly different but constant plate rotation speeds. Audacity just does not have this level of millisecond scale precision in its stock speed and tempo effects. This is a bug-level numerical precision problem but audacity bugzilla does not let new users to register, so I cannot report it to the developers directly.

Sounds promising, but I cannot do it manually and I do not have a plugin that can do it. Besides, high freq peaks must also be matched (at a later stage?) to get a full-spectrum record.

You are right about the sampling mismatch problems, but they are so minor in comparison with vynil scratches and pops engraved by ancient corundum needles, that they can be neglected at least at the first stage of vynil restoration.

I would like to have a tool like that but it MUST run on Linux. I am also reluctant to run closed source blobs. The cheaper version limits the length of aligned audio, so only the “Pro” version might suffice. It may have some weird protection scheme, so using this would amount to buying a new dedicated computer with an unfamiliar proprietary OS+the plugin for over 1500$ just to do a one-time hobby job.