Bitflipper's Top 10 Tips for Avoiding Artifacts with V-Vocal
There are plenty of legitimate gripes about V-Vocal, such as crashes. But there have been nearly as many gripes about how poor a job it does and how it leaves artifacts. V-Vocal has been responsible for a great many Melodyne sales.
Crashes, well only CW can remedy those. But artifacts, that's another story. If you follow these 10 suggestions V-Vocal will nearly always give you excellent results, free of audible artifacts. (These tips, BTW, apply to all offline pitch correction software such as Melodyne and AutoTune. Regardless of what you may have heard to the contrary
, no pitch editor in existence today is immune to artifacts!)
1. Always start with clean, moderately loud tracks If you have fan noise, rumble, headphone bleed or clipping, these will confuse V-Vocal. They won't necessarily ruin an otherwise great track, so if re-tracking isn't an option your best strategy may be to live with what you've got and skip the pitch correction altogether.
Fan noise, you might be able to reduce with Sound Forge, Adobe Audition or iZotope RX.
Rumble you can spot with a spectral display and remove with a high-pass filter.
Clipping may not be a problem if you simply go around it and don't try to pitch edit the clipped portions. This may be easy, because clipping is most likely on transients that don't need correction anyway (see tip #6).
But headphone bleed, that's a problem. Even if you silence it between phrases so that you can no longer hear it. Just because the main vocal masks it to your ear doesn't mean it can hide it from V-Vocal. It's still in there and it will confuse V-Vocal and will contribute more to artifacts than any other kind of noise. Reducing headphone bleed is addressed further in tip #10.
Also pay attention to recording levels. You want your levels to be moderately hot, peaking above -30db. V-Vocal employs an internal noise gate, so very quiet recordings may be partially or completely excluded from analysis, to the point where you may not be able to use V-Vocal at all on the track. If you're having problems with V-Vocal figuring out pitches (large areas have no horizontal white lines over them), try using Process -> Audio -> Gain to bring the track up in volume before starting your pitch editing. As long as the track is reasonably noise-free, this will work fine.
2. Never use V-Vocal on a stereo track There are few good reasons to record vocals in stereo anyhow. If you've done so accidentally, right-click on the clip and select "convert to mono" before applying pitch correction. (Or if you're feeling adventurous, split the track into two mono tracks and pitch-edit them independently. Beware of phase issues if you do this, however.)
A corollary to this tip: only work with monophonic (as opposed to polyphonic) material. Vocals are inherently monophonic, but V-Vocal can also be used on things like bass, and a bass track may not be completely monophonic if there is anyplace that two notes can be heard together.
Don't try to shift those places.
You can (sometimes) also fix a lead guitar solo, but only if you have a clean, dry direct recording and skip any sections where two notes overlap. (I have, however, seen V-Vocal used effectively on a lead guitar even though it was distorted and effected - but that was a single fix to a single flubbed note, and we probably just got lucky that time.)
3. Never use V-Vocal on effected tracks It's one more good reason not to print effects while tracking. If your singer needs to hear reverb, put it in the headphones but don't record it. Always record vocals dry if you want the option of applying pitch correction later.
This sometimes applies to natural reverberation, too. If your room is very reverberant, build some fiberglass panels or record in the closet where you hang your heavy winter coats. V-Vocal - or any other pitch editor - will work better without excessive room ambiance.
And if you have effect inserts on the track, bypass them while editing. They interfere with your own sense of pitch, and you want to be correcting by ear, not by eye. (See tip #5)
4. Never automate the correction process
Never allow any pitch-correction software to apply correction automatically (or in real time, as some plugins can do). Perform every edit manually, one phrase or word at a time, and audition every edit before moving on to the next phrase. You're not only listening to make sure you haven't introduced any artifacts, but also to determine whether or not your edit actually improved anything. Some edits can take a few tries to get it right, other edits are simply ill-advised and shouldn't be undertaken. But that's what the Eraser tool is for.
When auditioning, do so both with and without the full song playing along. Listening to the vocal soloed will help with fine details, but ultimately you're tuning the vocal to the instruments. It's no good achieving an on-pitch vocal if in doing so it's now no longer in tune with a key instrument behind it.
5. Do not edit things that don't need it Sounds obvious, but inexperienced users tend to want to fix everything whether it needs it or not. Even experienced users (including yours truly) will occasionally succumb to the lure of perfection and fix things unnecessarily.
Do not be fooled by the graphical display! Never assume that because that horizontal white line isn't exactly on the note that you actually need to fix it or that doing so will improve it. Listen to it in context
without looking at the display. Does it SOUND out of tune? If not, then leave it alone.
Background vocals generally benefit from gentle or no correction at all, especially thick multipart harmonies and/or double-tracked parts. Pitch variations are in fact critical to getting that desirable fat sound and you can actually ruin a backing vocal ensemble by over-correcting. Wobbly is
good when you're stacking human voices!
[Added:]
I should add that V-Vocal can add distortion even to portions of a clip where you have not made any adjustments, due to inaccuracies in the initial scan. They're not often audible, but they can be if your track contains difficult-to-analyze passages such as noisy and/or non-pitched bits. If this bites you, your best strategy is to split the phrase you want to edit from the rest of the clip and just edit that.
6. Do not edit non-pitched elements This is probably the most important tip of all.
Consonants do not have pitch, so it is impossible for software to pick out the fundamental frequency. Look at the spectrum of an "S" - it's basically white noise. What note is an "S" or an "F"? They have none. Attempting to shift consonants often results in nasty artifacts.
So when you select a region to pitch-shift, leave out the consonants. Don't shift by words, shift by vowels. Don't just grab the white lines that V-Vocal put there for you. Instead, specify your own edit region which doesn't include beginning or ending consonants. So, for example, if the word is "sat", highlight the "a" and leave the "s" and the "t" out of the selection.
Concentrate on long vowels, since they are the elements most likely to need correction. Most short sounds - less than an eight-note's duration - are too brief for a listener to register their pitch even if they have one, so many short vowels can be skipped just like the non-pitched elements.
[Amended:]
Tarsier points out that the blanket statement "consonants have no pitch" is not true, and he is right. The consonant "m" in particular can be drawn out to where it does indeed have pitch. It comes down to the length of the sound and whether it meets the minimum criteria for a listener to detect its pitch. A "t" never will, but an "m" might.
7. Do not edit breathy words, shouts, grunts, groans or screams Luckily, James Brown never needed pitch correction.
This is really a corollary to the previous tip: don't edit non-pitched elements. The same advice extends to pitched elements that have a lot of non-pitched components mixed in. These will make it harder for V-Vocal to separate the fundamental from everything else, and it will tend to try and shift components that shouldn't be shifted. The result is similar to what you get from excessive headphone bleed: phasey, metallic artifacts.
Especially beware of words that start strong but fizzle out into a breathy, amorphous cloud at the end. It may be better to skip such words altogether, since there is no clear delineation between the pitched and unpitched regions so you cannot reliably specify which parts not to shift. (BTW, this is one area where Melodyne really does excel over V-Vocal.)
8. Avoid abrupt or large pitch changes If you need to move a note more than a few cents, use the line drawing tool to insert a gradual change. Better to take a little longer to transition up to the right note than to immediately jump there and expose your edit to the listener. Don't worry that you've inserted an artificial-looking straight line, because you won't actually hear it as a straight line unless it's very long. Use your ears, and be sure to audition the edit before moving on, to make sure it's not noticeable.
As a general rule, don't expect V-Vocal to transparently move any note more than a semitone or so. If the vocal is that far off it may be better to insist on retracking it.
How well a big shift works depends a lot on if it's a high or low note (high notes can be low-passed to mitigate artifacts) and where it sits in the mix. I've manufactured high harmonies with V-Vocal that sounded absolutely dreadful when soloed, but worked just fine in the context of a multipart harmony in the full mix. (But it's hit-and-miss, so when possible I prefer to bring in a female vocalist to hit those high notes for me.)
This advice extends down to the micro level, too. Sometimes after editing one section you are left with a short but abrupt change going into the following section. This will result in an audible artifact. The vibrato tool can sometimes be used to match up the sections. This is one case where you have to do it by eye. If that doesn't work, use the line-drawing tool to draw a gentler transition. When matching up two sections this way, I try to terminate the line at the zero-crossing of the following waveform when possible.
9. Always set the Formant control to zero The formant-follow control determines how much pitch adjustment is applied to the non-pitched portion of the vocal. When you sing, there are fixed resonances in your mouth, throat and nose that do not change with pitch. You normally want V-Vocal to leave these alone, otherwise you'll get weird results such as the "chipmunk effect". Setting the formant control to zero tells V-Vocal not to mess with them.
The only time you need formant tracking is when you're synthesizing harmonies a fifth or more apart from the lead. Even then, expect to bury the faked harmony because it'll sound nasty in solo. And even when you do use the formant setting, you'll almost never set it to 100%, which by some strange logic is the default value.
10. Help the singer get it right the first time Of course, singing on pitch is ultimately the singer's responsibility, but there are many things you can do to help. And
every word or phrase sung correctly is a word or phrase that V-Vocal will never make worse (because you adhered to tip #5: don't fix it if it ain't broke!).
Singers who hear too much of themselves in the headphones tend to sing sharp. If they hear too little, they tend to sing flat. If it's too loud overall, they tend to sing flat. If there is too much bass, they tend to sing flat.
It's not enough to just give them their own headphone mixer that they can adjust. Listen to them while they sing - if they seem to be holding back and singing too quietly, it's possible their vocal is too hot in the cans. Conversely, if they're belting out to the point of losing control, they may not be hearing enough of themselves.
If you have an inexperienced singer, teach them the importance of backing off the mic. Use the pop filter as a barrier to enforce a no-eating policy. Not only will you have fewer issues with plosives, resonances and the proximity effect, it will also reduce headphone bleed, which will make subsequent pitch editing easier. If you need to bring the vocal up, try bringing the instruments down instead, to avoid increasing bleed.
The headphone mix can drastically help or hinder a vocalist, even one with perfect pitch. Too much bass will interfere with the singer's sense of pitch, often making them sing flat, so consider taking the bass and kick entirely out of the headphone mix, or at least rolling off the low frequencies with a HPF. Choose something (but just one thing) like an acoustic guitar or piano to bring forward in the mix to supply the main pitch reference to the singer. Lead instruments and incidental percussion usually don't help at all, so back them off or out.
A trick I use for my own singing is to insert a pitch cue right before the place I'll be singing a line. Just a note on a piano. This tends to mitigate that initial slide that sometimes happens when you're zeroing in on the first note. Those lead-in slides can be very hard to repair later and make double-tracking trickier.
Choosing the right headphones makes a difference. Most studios have lots of cheap headphones, because musicians are hard on them and they break often. But cheap headphones typically have poor isolation, leading to bleed, leading to pitch-correction artifacts. So give those disposable cans to the drummer and keep at least one good pair around just for vocalists.
Single-sided headphones work well for some singers because hearing their own natural voice in one ear helps with pitch. Beware the technique of simply removing one side of the headphones, though, as this will increase bleed unless you silence the unused side. There are models available that are specifically designed for this purpose, such as the single-speaker
Beyerdynamic DT102. A more cost-effective approach is to simply pan the headphone mix to one side.
The nicest-sounding headphones are not necessarily the best ones for vocal tracking. I like the Sennheiser HD280Pro partly because it has weak bass response, which saves me the trouble of high-pass filtering the headphone mix. It also has better isolation than most headphones (it even beats "active" noise-cancelling Bose headphones for isolation, so the HD280's are also usable for air travel). They're reasonably comfortable and durable, and best of all, cost only 99 bucks (or less).
I'm sure there are other good tips that I've either forgotten about or haven't even stumbled onto yet. Please add your own tips and observations, but please - no anecdotes about how V-Vocal crashes your system. Everybody here is sympathetic, but nobody can help you with that except the bakers.
post edited by bitflipper - 2011/04/08 10:50:04