Those home recorded choirs, bands and orchestras
In this lockdown there's loads of material on YouTube with brass bands, choirs, orchestra etc all singing at home and then it sounds and looks so simple.
I got asked - as a freebie of course, to do one for an amateur choir - you know the thing, really nice people but the prospect of using technology at home on their own daunting.
Clearly, my explanations were not working in text - so I did a video explaining what they had to do, and what I needed - which does seem to have worked.
The purpose of this topic is to explain where it went wrong, and how I sorted it, in case you get asked to do one and think it's simple - it isn't!
You'll get loads of submissions, and the first problem is you have no idea who they are, and many are random file names - 10293-48394.mov or choir video JW.mp4 really unhelpful things like that, so working out what it is is tricky.
My plan was simple - sync the backing track with the clip in Premiere, and then export just the audio as a .wav to put into Cubase. There is a backing track, plus an expected 36 (ended up slightly higher) voice tracks.
There were 6 sections - Soprano 1 and 2, Altos, Baritones 1 and 2 and Bass
I prepared 12 versions of the track to sing to to cater for people who were using earphone/headphones, one or two ears. So I did a mono mix, for people who wanted one ear open, and a split mix with track and low level other parts one side, and their part as a guide on the other. Some people liked one, others the other in a roughly 50/50 split. Quite a few are elderly and have hearing issues, so the two different tracks seemed to have been a good idea.
Phones provided varying quality - some clearly using a very low bit rate with that horrible computery sound, but others were very good. All a bit thin and they all needed EQ to boost the mid-range. A few were simply nasty sounding. Lots of people had page turn noise, doorbells and dogs barking, and 50% of the people didn't record the intro, and as the music plays out, many leaned forward and switched the recording off!
In cubase, I evolved a system. I had the guide vocal tracks each one of the six folders, and I imported each audio track and popped it into the folder.I had quite a few tries that worked badly, so I'll skip to the end process that worked.
Step 1. Normalise the track (I used -3dB).
Step 2. Variaudio analyse - a brilliant function in the latest versions of Cubase
For those who don't know it - Cubase analyses the audio file and produces and edit screen that has a keyboard top to bottom on the left, and over the top of a dimmed waveform, it shows the notes sung as coloured bars with the note name - so you see a coloured bar, saying G, but in this production, it might show as G sharp, or even A from some singer. In fact I ended up editing the pitch on every single note of every single singer. From experience, I know how easy it is to sing exactly wrong on every note because it's a phenomenon of having both ears bunged up with in-ear headphones. It's so simple to be persistently and accurately sharp or flat - so I was expecting that. BUT - some singers really could not pitch. The song they were doing was "Lovin that man of mine" and if you know this, the lyrics go "Love-in-that-man. . of mine" and in this arrangement the "lovin" should be a B natural. EVERY singer had got this wrong - many convinced it was C, but it's a minor chord there so the B was the right one. Cubase had C as a red bar, and B as a blue bar, and everybody was red.
So every track needed playing multiple times till it was in tune.
Step 3. Synchronise the phrasing. Everyone sung it differently - clearly without their conductor waving and mouthing the lyrics, it was rather, er, free. As in "till, I, die" or "Till-eye, dye" or "till, eye-die" - see the problem? So next step was to look at the guide vocal track visually, and cut up the track and place it under the guide vocal so they matched. This meant some notes were too long so they needed shortening with a quick fade out, and then the beginning of the next word might need a fade in, or not. Some would end up as cross-fades, but you have to play each two bar phrase in a loop and drag the tracks around till they suddenly sound right. The alto track had around 8 or 9 tracks - the busiest one, and chopping them all up and lining up the transients took a very long time.
I did this to each section, THEN discovered the guide vocals did not line up, so none of the sections lined up, so I had to align the guides, then go back to the sections and re-align everything. A time wasting mistake.
Once done - I colour coded each of the thirty odd tracks into excellent - dark green, through to red - terrible. One man really did sound like a Dalek. I then set up six groups and routed all the tracks to them. The groups had reverb added. Each individual channel had EQ applied. A few had dynamics applied to even out the quiet bits where the singers were a little unsure, against the pit they knew well and sung with gusto.
The colour coding helped with fader positions - all the green ones set higher than the red ones, that were kept low, but present in the mix - yellow and orange came somewhere between. I also made a not of those voices, that while accurate, were less nice to listen to, and they got kept lower to. I duplicated a couple of decent tracks, then did a slight re-arrangement of the cut up clips, and panned one either side of the centre to thicken up the baritone 2 section which was less in numbers.
Once the mix was roughly balanced, I exported that back to Premiere, and started to tweak the layout. With nearly 40 clips, all shrunk from their very random frame sizes to something common, and with each having a crop mask applied, Premiere was a bit laggy. So rendering out the clips was a necessity for smooth playback.
The sizing and moving them on the screen was a killer.
So far - I've got all bar a few tracks, and they will be here tomorrow, I'm promised.
In real terms this is probably around 30 hours of work so far.
Mainly revolving around identification - the woman with grey hair, the man with a beard, the man with a red shirt. I spent ages trying to find a cough in a ladies track and it was in one from another contributor. You also end up trying to cut out clicks pops and the computer sounds, but with so many tracks, they can drop out and come back in pretty much unheard.
The choir organiser has no realisation that one simple little change can take 4 hours to change and re-render, and will pick out tiny things and want them fixed when you cannot hear what the problem is.
My cunning plan to re-phrase the tracks with the guide tracks worked really well - but of course is making editing the video difficult because lip-sync fails. I've got to face having to cut a few video tracks and slip and slide them, so maybe to hide the gap, I'll move the little images around the screen?
I've learned quite a bit from doing this and 40 HD tracks, and FX on them is quite hard on the editor. Cubase doesn't get upset with all the pitch shifting - 40m tracks with every note edited means the audio folder is amazingly full of originals and copies, but all seems solid.
I think getting 40 out of tune and out of time tracks and fixing the audio has worked - I've put up a private link on YouTube if you want to see the edit so far - it's unfinished of course, but you can hear the audio. It started as a mess, but after the editing, I'm quite pleased.
nice job Paul
You did amazingly well there, Paul!
I just wonder about the time these things take, although another would be quicker now I know the mistakes. Like spending 3 hours on something and getting an email back - lovely, but I don't like the purple edges - so you scrap it and do a few more edits then they say, white would be good though - but that is a lot of edits back, so you have to dump something and recreate a new one from scratch.
Worst though are the phone video formats. iPhones seem to be pretty good, although the default seems to be 30fps. Premiere displays the specs of the clips and I was shocked to see really bizarre setups. With video, you normally expect 24,25, 29.97 and 30 as frame rates, but I had really odd ones - which did explain some of the jerky video and slipped sync - 16.34fps, or 11.5fps, or 30.025 - stuff like that. Media encoder can do much, but I did need too get a .3gp converter for some android file formats. This is for a good cause, so I think when it's done it will be fine - but the sheer amount of work was a shock.
I also think I should have had a spreadsheet that detailed the filename, the persons name and their description, and what part they were singing. I've ended up with loads of duplicate files, and moving between two computers with USB drives is mega slow in the workflow. It could have been good to change the filenames to something like John-w-row1-c3-bass, because one of the most frustrating things is finding the right track to adjust. Having to resort to turning off the view button and watching which picture disappears is a hopelessly frustrating feature - so being able to look for row and column and discover the right button to click without clicking them all one by one would have spared me so much grief. Exactly the same thing in Cubase - hearing a sniff, or a page turn then trying to actually find it takes such a long time. The idea to have each section balanced, like a multi-miked drum kit, then have a group fader to bring it up or down together is vital. Without groups, mixing would be terrible.
Another annoyance was that you needed to listen to multiple tracks with reverb to see if the edits were obvious, but for the time slipping you wanted reverb off, and that's a lot of clicking the mouse.
Oh yes - I tried using compressors to tame some of the loud to quiet dynamics changes but it brought up the background too much, so it's fader automation.
I also discovered the brilliant pitch correct system Cubase has is NOT effective in the bass range - sometimes, I could be a semi-tone adrift and the system would not let me put it right - I'd shift up and it was wrong, and shifting down was wrong - the right pitch being a bit elusive. With so many tracks on screen it's tempting to have them vertically narrow, but the pitch shift system switches between move and cut by looking where in the coloured band you are, and they're very close, so sometimes instead of a move, you'd get a cut.
Also - the Cubase stretch and match system I use for harmony vocals doesn't work very well on fixing the differences in phrasing. Normally I use this a lot, and you just set one track to be master, and then stack up alternative takes below it, and the timing of the master translates to the others and they get stretched and shrunk. Normally works well, but for some reason it wasn't as good as doing it any hand which is odd. It seemed to actually make it worse.
Finally done, but mixing at a distance is a killer. Wonderful, but at 2.32 the countermelody could do with being a little louder. Twenty seconds in cubase, a few seconds to export and FORTY minutes to re-render in Premiere, and then you get "sorry, you were right it's too much" and that's another 40 mins because you also tweaked a tiny thing they didn't notice!
If you do be of these, progress initially is fast but slows down to a crawl. It does show me cubase nowadays is much more comfy, computer wise. ALL computer issues were with premiere which was stretched badly with 40 video tracks, each one with movement, sizing and lots of effects. Cubase was 42 tracks of stereo audio all with pitch correction, dynamics, reverbs and cross faded two virtually identical computers. Premiere, at one point messed up some morphs I applied to jum cut edits and ten versions later it became unstable crashing randomly. I had to go back many versions till I found the problem and do it again. Cost me 6 hours!
You’re a braver man than I. And tip give anyone who ever wants to provide that sort of thing, handclap at start :), each and everyone.
Btw, sounds and looks fine on iPhone at least.
Normally I'd agree on the sync thing but ALL of this is visual - matching the music track with their first note because just getting it recorded needed a youtube instruction video to guide them, and so many were in and out of frame, or looking a totally different direction, or forgetting the words and doin just the one take. about half have been time stretched at the end because they just leaned forward and switched the phone off before the music finished. Normal music videos are simple compared to these. The other strange thing is that the size of the video files caused so much grief - drop box and similar failed badly because people had the option to put their music in one of six folders - the 6 parts, but most people put them in alto because it was the top folder. many didn't have their name or even the word alto in the filename, just what the phone produced - The two soprano parts are identical until one suddenly goes higher, and the other goes lower - you you have to listen before deciding. I did actually enjoy it, but had VASTLY under estimated the time required. As I progressed render time got towards an hour each time I made just a few changes.
@paulears Like I said, braver than me :). I had enough trouble wrangling 4 videos at once. In my defence though, very first time.
This is a tricky one because the average age is higher than the average, and technology is a bit tricky, and people got confused, so I sent them a link to a video.
If anyone is going to do this, I can share the videos they used to sing to, so you can hear what worked and what didn't. Just send me a message. I won't bore people who don't need them.
Now I've had another request to produce another similar project - small chamber orchestra and singers. Not sure this one will be so interesting? Even more interesting is how I will sort out errors in pitch, because it's going to feature Baroque music, which needs to be A=415Hz, not for any magical conspiracy theory reasons, but simply because they want to do it in period - so everything has to be slightly flat and that will mess up my Cubase pitch correct, unless anyone knows a way to change the way this works. My thoughts at the moment are record everything to modern pitch, then do the detune once it's done - any suggestions?
You're a brave soul, Paul...
I`ve been messing with Zoom and a few other of these conferencing tools. Not my cup of tea. Far too many bugs and workarounds. It could be simple, with WebRTC. But that wouldn't suit some, apparently.
I like you solution. KISS. Simple is the word.
Zoom is brilliant - for meetings and quizzes, but it totally unable to handle multiple streams in sync, although to be honest, that's the internet really - all the ever changing paths making it not latency, but real delays - so all these things are audio and video edits. I'm really mad with myself. I spent the afternoon playing in Verdi's Requiem - and I'm NOT a pianist, so it's slow, and finally got it in and nearly note perfect. Sent it to the organiser of the next one for approval, and she asked why I'd done Verdi - it was very hard? I apparently should have done a Vivaldi piece. She didn't know why I'd been sent the score for Verdi!!! I suppose they both began with V?
paulears, post: 464400, member: 47782 wrote: ...Sent it to the organiser of the next one for approval, and she asked why I'd done Verdi - it was very hard? I apparently should have done a Vivaldi piece. She didn't know why I'd been sent the score for Verdi!!! I suppose they both began with V?
No, it's much simpler than that - they both end in "i".