Ways to Reuse/Replicate a Vocalist
Derelict Dollhouse here, and if you’re familiar with that, you’re probably familiar with the biggest gimmick of my entire Udio use since the beginning: keeping the same vocalist. I can’t say for sure, but I think my catalogue is the biggest collection of purposefully vocally replicated work on Udio. That’s not a brag! That’s me going; the evidence of the methods I’m gonna go through is there! This article is going to cover how to replicate and reuse vocalists in a couple of ways, covering what I call my “old” method, and a newer one.
Replicating a Vocalist: The Old Method (Extension Method)
Approach One
1) Identify a song you wish to replicate the vocalist from, Identify a section of this song with about 30 seconds of the vocals you wish to use that includes the characteristics you want to reuse (highs, lows, etc) 2) Select “extend” for this song, and crop everything out except that 30 second section.
When you are cropping, have the “end” of the crop window cut into vocals (so the vocalist is interrupted mid delivery). This helps promote fresh melody start.
Set up the core prompt for this extension for your desired song genre, ensure you do not use any prompts that steer or specify vocals
Select “extend after” and select “Instrumental” OR in the custom lyrics box, type a lyrical prompt that would command an instrumental.
In the advanced options, set the song position slider to 0 (beginning), and the song context slider to 1 second. The rest of the settings you may adjust as to your personal preferences. When this is done. Hit extend. 3) The generations should spawn with a “new” melody. This stage is now a test of patience of scouting for a new melody you find interesting for a new song. So these clips will be 30 secs of old vocals, followed by an abrupt shift into a new song intro. We start with an intro to force a melody break, you can always extend the intro into a further part of the song instrumentally if you like. 4) When you identify a clip you like and wish to work on, click “extend”.
Return to the core prompt and re-enter vocal prompts as they were in the original clip you were extending from. Leave all other prompts as they are or change them as you wish the song to progress from the introduction clip you have.
In the lyrics box, select “automatic” or “custom” as to your preferences. Type lyrics into the box if doing custom. Custom is probably better here because you can purposefully make the lyrical cadence of the “new song” different from the old one, which helps our attempts to “move” the vocalist into the new melody, rather than have the old melody return (which we don’t want).
When lyrics are finished, adjust song position slider as you see fit, and for song context this is where some additional considerations come into play:
The more of the “old song” Udio can see, the higher the chance it will faithfully replicate the old vocalist precisely HOWEVER it also increases the chance it will be influenced by the old melody.
The less it can see, the more room for variations in vocalist replication you will have, but it is less likely to copy the old melody.
You need to select an amount of context for Udio to see (at least 4 seconds of the old vocals must be visible) that fits your requirements.
Some tips for this:
The closer your new song genre is to the “old one” you’re extending from, the higher the chances of melody replication. So you want less context windows if you’re doing this. In my case, if I’m making a new song with an identical core prompt to the old one, I will generally have my context window allow Udio to see about 5 seconds of “old vocals”.
If the genre is distinct, you can allow Udio to see more of the old vocals, as the direction for the new generation will be connected to your current prompts which increases the chance the vocalist will be “pushed into” that musical flow. For example when working with a brand new prompt on a new song, I’ve managed to replicated a vocalist even when allowing Udio to see 90+ seconds of “the old song” and still got the vocalist singing to the new melody.
NOTE when selecting your context slider you must remember the instrumental you’ve generated (or however much of it you wish to use) needs to be factored in, so this MUST be added to the context window. So for example if I’m going to use all of the generated instrumental intro (32 seconds) and want Udio to “see” 8 seconds of my old vocalist, context slider would need to be set to 32+8, or 40.
When context slider is set up, hit extend. 5) Now all you need to do is parse the clips to listen whether the vocalist comes in on the new melody. This can take some resilience, particularly as Udio can make a singular voice produce some different timbres (especially if you’re switching genres) so it’s very much trial and error. Once you have a clip you’re happy with, in the next extension you can crop out the “old song” pieces and use your new vocal clip in the new melody to begin work on the new song (or simply continue it).
Approach Two
1) Identify a song you wish to replicate the vocalist from, Identify a section of this song with about 30 seconds of the vocals you wish to use that includes the characteristics you want to reuse (highs, lows, etc) 2) Select “extend” for this song, Crop the song so the end of the 30 seconds you’re interested in is the start of the new clip. This will work better if you do not “cut” vocals in half and have a pause when the new part starts.
Setup your prompt to your intended genre, leave the vocalist tags you have in the old song alone
In the lyrics section, select “custom”, you’re going to write lyrics that differ as much as possible from the lyrics in the prior section in terms of metatags and also structure. This will encourage Udio to “shake up” the melody, hopefully enough to give us a new building block without having to gen an instrumental to lead it prior.
Some tips for this:
Whatever the prior section metatag is, differ your new one. If it’s a verse, write a chorus or bridge. If it’s a chorus, make it a verse or a breakdown. Bridges and breakdowns are very good for denoting a melody shift so they’re good ones to use to give you “new melodic clay” to work with.
Look at your rhyming scheme and vary it. If you used AABB in the old section, do not use AABB in the new one, so the melody “doesn’t fit”. Try something like ABACCA. Vary the line numbers and lengths as well. If your previous section had no repetition, consider using some.
The idea behind this isn’t necessarily to generate a “workable” vocal clip first time around, it’s to shift Udio off the old melody onto a new one, with the same voice so you can crop out the old one and start building from there. Similar to the previous approach but potentially quicker because you get workable vocals in your first gen. 3) As with the prior method, you need to be mindful of your song position slider (vary it from the slider position of the “old clip”) and the song context slider, allowing Udio to see “enough” of the old song to catch the vocalist but not too much so it just “carries on with the same song”, it depends how varied in structure your new lyrical section is. The more varied, the more you can allow Udio to “see”.
When these options are set up, hit “extend” and then it’s a case of parsing the clips for value as with the previous method, and finding one you can “crop and tidy” of old material and begin work on.
Replicating a Vocalist: The New Method (Styles)
1) Identify a song you wish to replicate the vocalist from, Identify a section of this song with the vocals you wish to use that includes the characteristics you want to reuse (highs, lows, etc) and styles this song and portion of song to generate a new fresh clip that includes vocals. 2) Set the “similarity” slider to maximum. This is incredibly important. 3) Make sure in the “styles guidance” box, your vocalist tags match those used in the song you are styling from. You can vary the others much as you like. 4) Tweak other options as to your preferences. When scouting initially, you want to select for a 32 second generation as they’re quicker, as we need to identify ”the sweetspot” for our song. Use auto lyrics as that’s quicker too.
What is the Sweetspot?
The sweetspot is the name I give to the styles selection area we use that faithfully replicates the vocalist accurately. This can take some time to work out. Even minor slides of a second either way can make a styles replication go from perfect replication to “sounds similar but not quite”. Identifying a sweetspot can be credit exhaustive. When doing so I’d recommend running at least 4 gens, listening, and if they’re not right, slide the styles window slightly and try again. It took me many credits to work out the sweetspots for several songs, so have patience. Some vocalists appear to be stronger replications than others depending on the song. For example where I use my main vocalist (F) in a specific song sweetspot; her replication rate is essentially 100%. Every gen will be her without fail. With my second vocalist (E) the best sweetspots I’ve found only have a capture rate of about 15% or so. This is possibly due to her voice being a “foreign implant” into the genre I work in originally.
My point is not all songs and vocalists will behave the same, but through this method so far I’ve managed to replicate 7 different vocalists now, across a few different genres (including placing vocalists in odd genres, and even dueting some of them together). So I feel it can work for most if not all vocalists, you just need to work out the sweetspot.
Once you think you’ve found a sweetspot, test it a few more times to ensure it is giving what you want. If it returns the vocalist, make a note of the position on the styles context window. You will not want to forget this!
With a sweetspot identified, you can now try generating “proper songs” with proper lyrics and even 2:10 length if you like. If your identification of a sweetspot has been correct, so long as you do not alter the vocal “styles guidance” tags or similarity, you should get the vocalist replicated.
🧠 Contributor: Derelict Dollhouse
Last updated