Give your vocals a completely new personality with IK’s ReSing.

Vocal synthesis, whether for sung vocals or spoken‑word dialogue, is undoubtedly a process that has made massive leaps forward in recent years, underpinned by advances in AI‑based processing. From the perspective of music/audio production, these products broadly fall into one of two categories: virtual singers (virtual instruments that can generate sung vocals) and voice changers (processors that transform a source voice/vocal into a different voice/vocal). The subject of this review — IK Multimedia’s ReSing — sits within this ‘voice changer’ category. So, if you have a single voice source for sung or spoken vocals, and want to change the character or gender of that voice, or even transform it into a musical instrument, just what can ReSing offer?

The Small Print

Having already implied that ReSing leans into AI, let’s head off the most obvious (and perfectly legitimate) elephant that might be in your room. IKM make it very clear within their description of ReSing that any training materials used within the product — in this case, the voices of specific singers or voiceover artists — have been used with suitable permission from (and compensation for) those individuals. Assuming you have the rights to use the original source recording you are transforming, any use of ReSing’s voice‑changing technology should not, therefore, get you into any ethical or legal hot water.

A second important initial observation is worth making. Sitting within the ‘voice changer’ category, ReSing can transform your sung vocals so that they sound like they were sung by somebody else. As a result, while it can change the character of the vocal delivery in some useful and interesting ways, any dubious pitching within the original source vocal is faithfully recreated in the transformed version. ReSing should not really be considered as a magic button to turn a bad singing performance into a great one.

ReSing should not really be considered as a magic button to turn a bad singing performance into a great one.

The Voice

ReSing is available as a single‑purchase perpetual product (no subscription required) and once installed, all processing is done locally. The standard version includes 10 different voice models, 10 instrument sounds and 10 ‘model generations’ while the Max version ups all three numbers to 25. Yes, voice‑to‑voice transformations aside, ReSing can transform your vocal source material into the sound of selected instruments. Equally, if you want to train your own voice models, ReSing supports that. There is an additional ‘unlimited model generation’ add‑on you can purchase if you need to train more voice models beyond the included 10 or 25.

Once installed, ReSing itself can run as a standalone application, a conventional VST, AU or AAX plug‑in or, DAW permitting, via ARA. The latter provides a more streamlined workflow, but both of the other options are perfectly workable; you just drag and drop the source audio into ReSing and processed versions are simply rendered to your drive system.

Personality Permutations

ReSing’s operation is very straightforward. A model browser lets you load up to two models (voices or instruments) into the Model A and Model B slots. Once your source audio has been imported, three circular control sections — Character, Accent and Dynamic — allow you to adjust how either one of these models (or a blend of both of them using the Fusion sliders) are applied when you hit the Process button. If you make subsequent adjustments to these controls, you have to re‑render the modelling. Thankfully, on any recent computer system, processing is actually quite speedy, and much faster than real time. The degree of control offered by the various Character, Accent and Dynamic settings is somewhat source and model dependent and you perhaps shouldn’t expect to transform a whispered original into a raging rock monster by simply choosing the ‘Energetic’ Character flavour, but you can subtly nudge the original performance in the required direction by experimenting with these options.

The fourth circular control section provides compact takes on dynamics, EQ, reverb and automatic pitch correction. Applied after the transformation process, these are neat to have, particularly if you are just working in the standalone version of ReSing. However, you can easily bypass them when working within your DAW if you have preferred plug‑in options for these tasks.

I have to say that the results of the voice transformation process itself are very impressive. For example, when starting with well‑recorded sung vocals, even when the source was entirely different from the model (different gender, different vocal range, different vocal character), more often than not, the transformed vocal could easily be dropped into a mix without its synthetic origins being obvious. It’s worth noting that I did manage to get some decent results from vocals that were originally sung into my phone, and ReSing generally added a bit of a polish to that recording once the fledgeling idea was inside my DAW. That said, the cleaner the original, the less likely you are to experience processing artefacts.

Max’s 25 voice models provide a broad palette of very useful vocal styles, from the gruffness of Spencer or Neil to the polished pop‑friendly flavours of Clara or Gabriella (the latter with a rather nice hint of a Spanish accent). Equally, there are models that could work very well with rap‑style vocals or for voiceover transformation. There are even a few special character options including the cartoonish Emily, Chipmunk and Synth Choir. It’s not difficult to see how sound designers might find a tool like ReSing useful for creating voice‑based special effects.

The various instrument models are also quite interesting. Yes, they do require a little more careful handling in operation if your intention is to achieve a ‘natural’ result, but if you vocalise an appropriate melodic line in your best impression of the target instrument, the transformed results can be rather good. For example, I had some success with the two saxophone and the trumpet models, transforming short vocal phrases into instrumental parts. And, of course, there are some weird — and sometimes wonderful — creative options to explore here. This AI‑based approach to virtual instrument sound creation is perhaps still in its infancy, but the potential is both obvious and very interesting. Watch this space.

Model Construction

ReSing also lets you build your own voice models. Done via a separate application, the principle is simple; you just drop audio recordings from your model target into the engine, and it then analyses these to ‘train’ the model.

The PDF documentation provides guidance on the type and quality of source material required but if you wanted to do this to the standard of IKM’s included models, I suspect it would require a fairly rigorous approach to compiling the source material, spanning expressive singing covering the full pitch range and energy levels within the singer’s vocal repertoire, sustained vowel sounds, syllable‑based scales and examples that emphasise sibilants and fricatives.

Bearing in mind the limited model generation counts offered by the standard and Max editions, it’s perhaps something you would be wise to plan carefully. Equally, the analysis of the training data can take a considerable amount of time depending upon your computer hardware. I did have a stab at this process, albeit with a sub‑optimal source data set, and the results were interesting if a long way from the quality of the included models. Again, the potential is clear to see, and I can easily imagine film/TV dialogue replacement suites experimenting with ReSing models trained on lead actors within a project to complement (or even replace?) the traditional studio sessions often required for on‑set dialogue replacement. Incidentally, ReSing does allow you to import models created by others in both ReSing and RVC (Retrieval‑based Voice Conversion) formats.

Is ReSing The Thing?

In terms of competing products, Sonarworks’ SoundID VoiceAI is the most obvious alternative to ReSing. This is also available as a one‑time purchase perpetual licence, and offers both voice and instrument transformation options. It doesn’t provide model training capabilities, but it does include a rather neat ‘unison mode’ for generating double‑tracked voices. Having run both side‑by‑side, they produce comparable results so a personal choice might need to be based on the broader feature sets and pricing of the two products.

It can provide you with endless voice identity options if your lead vocal, backing vocals or dialogue need to be delivered in a different vocal timbre.

I’ll finish with a reminder that ReSing (and, indeed, VoiceAI) is a tool for voice character transformation and not a software solution for rescuing technically poor singing by magically resynthesizing it into technically great singing. That said, it can certainly add some delivery polish to a well sung but scratch recorded vocal and, equally, can provide you with endless voice identity options if your lead vocal, backing vocals or dialogue need to be delivered in a different vocal timbre.

Is swapping out the character of one voice for the character of another voice any different from picking a different virtual drum kit or orchestral string preset (things many of us accept doing without question) to better suit the needs of a production? I’ll leave you with your own ponderings on that front but, if it’s a concept you are comfortable embracing, IK Multimedia’s ReSing provides the technology required to do just that.