By John Walden
The initial beta release of Omnivocal ships with two voice characters.
Find out just what’s possible — and what’s not — with Cubase’s new virtual vocalist.
Steinberg dropped something of a leftfield surprise with Cubase 15: the inclusion of Yamaha’s Omnivocal. Supplied initially in a beta form but available to Pro, Artist and Elements users, Omnivocal is a virtual vocalist instrument plug‑in. Of course, Yamaha have a longstanding history in vocal synthesis — for example, it was way back in March 2004 that we first reviewed their Vocaloid software, whose synthetic vocals have spawned something of their own genre online, particularly within the anime/manga world. But in recent years, developers such as Dreamtonics and ACE Studio have used AI techniques to push vocal synthesis technology forwards, allowing users to craft remarkably natural‑sounding and expressive vocal lines from a virtual instrument in much the same way as they might for a virtual string or wind instrument.
On the surface, it would appear that Omnivocal aims to do something similar — in essence, you just instantiate it on an instrument track, record (or program) a standard MIDI clip with the desired melody, and Omnivocal will then synthesize a human vocal to ‘sing’ those notes. So for those who’ve upgraded to Cubase 15 or are considering it, let’s find out exactly what you can expect from your new virtual session singer...
Singing Lessons
As with other virtual instruments, you can choose between different sounding presets. In its current beta form Omnivocal lets you choose between a male or a female singer, but I gather more options featuring different singing characters are planned for the future. You also get a number of parameters that can adjust the sound character of each singer. For example, you can pick between Straight (more subdued) and Dynamic (more expressive) singing styles.
There are also automatable parameters, such as Formant, Attack, Air, Power and Vibrato, all of which can change the character of the vocal delivery. You can automate these parameters, but they require the synthesis engine to re‑render the sung vocal to reflect your changes. While this happens automatically and very quickly in the background, it does mean that you can’t actually adjust these controls in real time like you might with, for example, a synth’s filter cutoff. That said, pitch‑bend data and the Presence (a tonal control) and Output dials operate outside the vocal synthesis engine, so they can be adjusted in real time.
The Text box allows you to add lyrics for Omnivocal to sing.
Initially, your melodic line will be sung with an ‘ah’ vowel sound for each note, but you can enter your own lyrics in the Text box on the MIDI Editor’s Info line. It has to be said that, in its current beta form, this aspect of Omnivocal feels a little cumbersome in operation. You can enter lyrics one note at a time, but you can also enter multiple words. For example, entering ‘broken heart’ into the Text box of the first note of a three‑note phrase would spread the words on a one‑syllable‑per‑note basis. So, the syllables of ‘broken’ would be placed into the first two notes (with the ‘oh’ sound being sustained across the two notes) and ‘heart’ into the third note.
Helpfully, once Omnivocal has analysed your lyrics, the phonetic form is also shown in the Text box and superimposed on the MIDI notes. In addition, you can enter some special characters, the most important being a hyphen (‘‑’), which forces the engine to extend a vowel sound across multiple MIDI notes. You can also customise the phonemes used in the Text box (the Omnivocal PDF manual has useful information on this) but learning how to finesse the pronunciation of your lyrics is something that will require a little trial‑and‑error experimentation.
That caveat aside, with your notes and lyrics added, Omnivocal will apply its AI‑assisted engine (all the processing is done locally and based on a voicebank of data for each virtual vocalist) to render the vocal part and, on playback, it is then performed with the rest of your project. And, of course, as with any virtual instrument, you can then change the performance by editing the MIDI note pitch, timing and lengths, changing the lyrics, or adjusting the vocal character as required.
Getting Hooked
Whether it’s in the writing/composing phase, or as a key production element in a top‑tier final mix, we’re now all very accustomed to the concept of virtual instruments being an integral part of the music that we hear. So, in that context, where can this initial Omnivocal beta release get us? Is it useful only as a writing tool (a less ambitious bar to jump) or can it deliver a final performance (a much higher bar)?
To answer that question, I’ve worked through a few examples, and you can find some audio clips to accompany them on the SOS website (https://sosm.ag/cubase-0226). I started with a simple target in my sights: creating a vocal hook or two that might be used in an electronic music project. The screenshot with the MIDI clip shows an example of what’s required for this sort of thing. I played the melody in from my MIDI keyboard (Omnivocal provides a simple synth sound for auditioning purposes as you play), over the top of a suitable backing track. I then entered the lyrical content in the Text box, one phrase at a time, and Omnivocal mapped each phrase across the appropriate number of notes based on the syllable count.
This simple vocal hook idea was more rhythmic than melodic, and once the lyrics were rendered by the engine I did some tweaking of note timings and lengths to improve the flow of the vocal delivery. I also picked the Dynamic delivery mode, and set both Attack and Power quite high to provide a more forceful performance style. But while there was some tweaking involved, the initial process was remarkably quick: just a few minutes’ work. As demonstrated in the audio example, I then generated a number of different variants with some ear‑candy processing applied to spice things up. I’ll make no great claims for the composition(!) but in sonic terms at least I feel that Omnivocal is a perfectly capable tool for experimenting with these kinds of vocal ideas.
As seen for our backing vocal example, once you have created one Omnivocal part, as with any MIDI‑based virtual instrument, it’s easy to then create – and edit — multiple variations.
Back Me Up
If you already have a human lead vocal in your project, a second possible use case for Omnivocal would be to experiment with potential harmony and/or backing vocal parts. Again, you can create the required Omnivocal MIDI by playing along to the lead vocal part using your MIDI keyboard. But I adopted a different approach for this example, instead applying Cubase Pro’s VariAudio to my lead vocal, then exporting the extracted MIDI note data. This MIDI data was then imported into my initial Omnivocal track for preparatory editing and the addition of the lyrical data.
It’s at this point that the ‘utility’ nature of Omnivocal comes to the fore, because whatever type of backing vocal part you want to experiment with (double, octave, third‑above or counter‑melody, say), it’s now easily explored with MIDI edits. And if you want multiple backing lines, then just duplicate the first and edit the notes. I explored precisely this workflow to create the second audio example, going from some simple doubles to a pair of basic third‑above harmonies, and then finally to something closer to a counter melody... and all with the ease of MIDI editing, which included some subtle randomisation of the note start positions for each harmony part. Whether Omnivocal’s backing lines make it to your final mix or not is another matter, but it certainly makes it easy to explore ideas for backing‑vocal parts in the absence of an actual singer.
Take My Lead
By this stage, it shouldn’t take much of a leap for you to start wondering if Omnivocal could be used to generate lead vocal demos as a guide for an actual human vocalist. Well, it can, and the basic process remains the same. As with the backing vocal example, once you have an initial Omnivocal performance, the simplicity with which you can edit the timing, pitch, lyrics and, to a certain degree, the performance style, make it a really useful songwriting tool.
That just leaves one obvious and very big question: can Omnivocal deliver lead vocals that are good enough to use in a final mix? Well, context is everything here, of course, and Omnivocal is still in beta, so expecting it to jump over the highest of realism bars at this point is perhaps unrealistic. Right now, I think the answer is ‘probably not’! For fun, though, my final audio example is a comparison between an Omnivocal lead vocal and the same vocal part created using Dreamtonics’ Synth V. Currently, the most obvious challenges when using Omnivocal are in crafting notes where words flow across multiple pitches. There are also limitations when it comes to adjusting the length of and emphasis placed upon particular syllables to influence pronunciation and delivery style.
But fingers crossed, there is more to come, and in any case don’t let any of that stop you exploiting Ominvocal’s considerable potential as a vocal sketching tool. Hats off to Yamaha/Steinberg for bundling this fledgling virtual singer instrument with every main edition of Cubase 15: whether you’re a Pro, Artist or Elements user, if you can hum it (or rather, play it on a MIDI keyboard), Omnivocal can sing it.
