Transferring multitrack cassette music to Pro Tools

[yonkiman]

Originally Posted by sw rec

Trust me, I get it. I've been doing this for a VERY long time. You can transfer a set of tracks from an analog machine, then transfer the same tracks from the same machine and they won't be in sync. This isn't guessing going on here, this is TONS of experience.

OK, for the benefit of everyone reading this thread (if not you personally, sw rec), what I'm talking about is flat out SOLVING the problems you are describing. I completely understand the problems you are talking about and the difficulties, that is why I want to find a solution.

Yes, in the old days it was impossible. Sure, you could do heroic things like riding the pitch control as cwsand suggests (my hat's off to anyone that could pull that off!). You could spend lots of time cutting or inserting a few milliseconds here and there throughout the tracks (super-time consuming but with enough effort you might adequately get through an 8 track song in a day or two). But basically no matter how many TONS of experience you have it was near impossible to do.

Let me try to put it another way (thanks to cwsand for the analogy): What I'm basically talking about is having the computer ride the pitch control. Look at Autotune and Melodyne and you get the idea. The tracks from tape 1 are the reference "pitch". When the tracks on tape 2 are a bit sharp relative to 1, you resample the track 2 data to lower the pitch proportionately. When the tracks on tape 2 are a bit flat relative to 1, you resample the track 2 data to increase the pitch proportionately. It's like Autotune and Melodyne, except the changes needed to the sample data are so "holistic" and incremental that there would be no noticeable artifacts. And you don't have to look at the relative pitch, you could also look at the relative transient timing (like Beat Detective).

It's not rocket science, but thanks for spurring me on into doing it myself! I'll probably have a working prototype just in time for Antares to release a commercial version!

[cwsand]

Originally Posted by yonkiman

Let me try to put it another way (thanks to cwsand for the analogy): What I'm basically talking about is having the computer ride the pitch control. Look at Autotune and Melodyne and you get the idea. The tracks from tape 1 are the reference "pitch". When the tracks on tape 2 are a bit sharp relative to 1, you resample the track 2 data to lower the pitch proportionately. When the tracks on tape 2 are a bit flat relative to 1, you resample the track 2 data to increase the pitch proportionately. It's like Autotune and Melodyne, except the changes needed to the sample data are so "holistic" and incremental that there would be no noticeable artifacts. And you don't have to look at the relative pitch, you could also look at the relative transient timing (like Beat Detective).

Yeah, it works fairly well though they can still get out of sync a bit in between the beginning and the end of the audio. The shorter the section you're trying to sync, the better it works. Of course, you could even throw Elastic Audio and Pitch in there for another fun nightmare!

[Carl Kolchak]

Whilst there is no automatic way of re-syncing your tapes, I've used the following method with great success transferring from an old 144 into PT.

Note : this is quicker and easier with the 144 rather than the Tascam 424 because in addition to the stereo outputs, it has a discrete tape cue output, meaning you can transfer three tracks at a time - allowing for a phase coherent stereo track + a mono track or guide track.

1.

a) If your drums are recorded in stereo, transfer them first (to a stereo track).

b) If your drums are mono, send them out of the left channel and send a second instrument out of the right channel, recording to two mono tracks in PT.

Your second instrument should be the one that it would be least acceptable to perform edits on (usually things like vocals and bass have plenty of space to allow for edits to be invisible).

2. Now transfer all of your other tracks, always sending the drums out the left channel and your second instrument out the right, and recording to two mono tracks.

Always group each new pass, so you have drums + instrument.

3. When you have finished all the transfers, click in your original drum transfer and use "Tab to Transient" to find the first beat.

4. Drop a sync point.

5. click in the drum track of your second transfer pass and use "Tab to Transient" to find the first beat.

6. hit "B" to separate the regions (because you grouped the drums and the instrument, the cut will apply across both).

7. repeat for subsequent transfer passes.

8. now click in your original drum transfer and use "Tab to Transient" to find the sync point.

9. right click on the newly separated drum region from your second transfer pass and choose snap to selection.

10. repeat for subsequent transfer passes.

11. repeat steps 3 to 10 every bar or so (you may find in some places you need to do it every few beats, and in others, not for several bars).

12. now you can ungroup and deactivate all of your drum guide tracks from the additional transfer passes, hide them and get to work smoothing out all the edits in your other instrument tracks - this may include addressing some pitch shift issues on a region by region basis, but in my experience, that hasn't really been a big deal.

Having said that, I wasn't trying to reconstruct a song from 3 or 4 separate cassettes, so you'll have to suck it and see.

It's actually a lot quicker than it sounds - and may well be even easier if you were to use elastic audio instead of hitting "B", but I've never tried it that way as I didn't want to introduce any artefacts.

Hope that helps.

[yonkiman]

Originally Posted by Carl Kolchak

It's actually a lot quicker than it sounds - and may well be even easier if you were to use elastic audio instead of hitting "B", but I've never tried it that way as I didn't want to introduce any artefacts.

Thanks for the detailed suggestion! I'm not at home, so I can't try this out or completely visualize what you're saying. I understand using the drum track as the marker - and that should work well (when there's a drum track - but I'm not sure how you're re-aligning the drifting tracks. I could see using elastic audio to stretch/shrink every bar as needed to align with the master drum track, but I don't understand how you're doing it without elastic audio. Are you just moving samples back and forth in time? It seems like that would create gaps/overlaps.

It may all make sense when I'm in front of the PC with enough time to time to try it (probably this weekend).

Thanks!

-Fred

BTW, "Kolchak: The Night Stalker" was my favorite TV show when I was a kid. Don't know if you enjoy or hate the association, but I thought getting a response from Carl Kolchak was pretty cool.

[Carl Kolchak]

Originally Posted by yonkiman

but I'm not sure how you're re-aligning the drifting tracks. I could see using elastic audio to stretch/shrink every bar as needed to align with the master drum track, but I don't understand how you're doing it without elastic audio. Are you just moving samples back and forth in time? It seems like that would create gaps/overlaps.

Each time you do step # 9 you are snapping the newly cut region back into sync with your guide track - sometimes it's early, but mostly it's late.

When you come back and clean up the edits, you might want to drag the beginning of a region out to reveal the start of a note or lyric for example ( as it stands the cut is effectively quantized to the beat). It's very simple to adjust the region boundaries to create clean transitions and remove any double notes etc - worst case scenario is that you might have to find a moment of "air" to paste over something.

Most of these edits will be invisible in isolation, certainly in the context of the final mix.

Originally Posted by yonkiman

BTW, "Kolchak: The Night Stalker" was my favorite TV show when I was a kid. Don't know if you enjoy or hate the association, but I thought getting a response from Carl Kolchak was pretty cool.

Ha Ha! Busted!

Yeah, I never use my real name at all on the net (not even on IMDB - which in the cases of some of the trash I've worked is certainly a good thing).

When I first joined this forum the vast majority of the members were using a nom de plume, so I just followed suit - it was either going to be Jim Rockford or Carl Kolchak - both childhood faves (see how James Garner totally kicks Bruce Lee's ass in "Marlowe"), and I figured CK was the more obscure and might make somebody chuckle one day....

...congratulations, you're the first!

Actually I really regretted it when they did that truly godawful remake series a few years back, the only thing more cheesy would have been to call myself Fox Mulder - truly lame!

[mr.cello]

Here we are, 7 years later and I am trying to do exactly what yonkiman was trying to do originally. Was it successful? Are there tools/plugins that will help do this? It has been seven years

[yonkiman]

Originally Posted by mr.cello

Here we are, 7 years later and I am trying to do exactly what yonkiman was trying to do originally. Was it successful? Are there tools/plugins that will help do this? It has been seven years

Ha! Well I have good news and bad news. The good news is that it definitely can be done for most music with some sort of clear beat. I found a python module that takes a sound file, sends it to a server somewhere, and returns the timing of every beat. I got as far as sending it song segments and generating song tempo maps that show how the tempo slowly varies across the length of the song.

The bad news is that once I prove to myself something challenging can be done, I lose an enormous amount of the energy I need to finish the job (the other 90% of the work). I should have been a professor.

IIRC, the two things left were:

• A fairly simple non-linear filter for the tempo data. The module occasional sends an erroneous measurement, particularly if there's little or no percussion. But since the tempo varies so very slowly, it would be fairly easy to detect an outlier and interpolate using the previous and next measurements.

• The sample interpolation. You need to re-sample each measure of track B to make it match track A. Again, a fairly simple linear interpolation would probably work well enough here, though a more sophisticated resampler would ensure no artifacts.

I've been meaning to get back to finish this for (apparently) 7 years. Maybe this is the nudge I need. But I wouldn't count on it. So if any programmers out there want to tackle finishing it, I'm happy to share my existing python code for generating the tempo maps (as long as I get a copy of the finished product).

Interesting aside: The tempo extraction module was not local - a call to it sent your audio sample to a server somewhere on the internet where it was processed and then the map was returned. I don't know for sure if that exact service is still available, but in 7 years it's possible that something even better has come along, possibly something local to your PC.