Baby Audio Humanoid

Transform your boring old human vocals with Baby Audio’s latest plug‑in!

Mangled, manipulated vocal sounds can be a big part of many electronic, dance and experimental musical genres. And they’re exactly what Humanoid, a new plug‑in from California‑based Baby Audio, is all about: it aims to be a one‑stop shop for a range of really synthetic, processed treatments, including hard tuning.

Plug‑in formats are VST, VST3, AU and AAX, and you’ll need to be running at least macOS 10.11 or Windows 10. Humanoid is an audio effect plug‑in with an optional MIDI input, so most serious DAWs will be able to access all its features. The interface and style seems ideally suited to the iPad too, but there’s no word on a version for that OS just yet.

Human Behaviour

Fire up the plug‑in and you’re met with a modern, continuously resizeable single‑window design, and the five user interface panels give a strong indication of what effects are on offer.

Pitch is Humanoid’s take on Auto‑Tune and works in a few different ways. Scale mode is conventional, detecting the pitch of incoming monophonic audio and bending it to a quantised scale degree. Preset scales and an interactive keyboard graphic are provided to let you choose your desired notes. In Note mode pitch is constrained to a single pitch drone, or in Duo mode a pair in parallel, across a five‑octave range. MIDI mode lets you route into the plug‑in live or DAW track‑based MIDI data to ‘play’ the pitch correction, and it’s five‑note polyphonic too, making for a vocoder/harmoniser‑like experience. MIDI pitch‑bend is respected, to the tune of ±2 semitones, but neither velocity nor aftertouch (let alone MPE) is observed, which is a shame. Finally a Lock option lets you maintain settings here even when exploring other presets.

Under‑the‑hood ‘calibration’ parameters, shown here for the Robotify knob, allow for a deeper dive when required.For each mode there’s a small handful of parameters, but it’s worth noting immediately that pitch correction is always ‘on’ even when all controls are turned down. The Quantize knob would then seem to be a sort of macro, adding artificiality by increasing retuning speed and clamping down on pitch fluctuations. Nearby there’s a familiar Formant twister, and also an intriguing ‘Robotify’ knob. This conforms the harmonic series of the input signal to a mathematical ideal, but also boosts upper harmonics to create an unnatural but not unpleasant plastic, zingy, almost cute quality.

Throughout Humanoid more nuanced control is available for parameters that have a ‘cog’ button: clicking one in the Pitch section for example accesses up to four ‘calibration options’, which are really additional parameters in their own right. So there’s a nice balance here of immediacy and depth: you can do a lot of damage with just a few mouse drags, or spend longer to really explore the effect. The same is true of the Utility panel, that limits pitch correction range, has a useful, dry‑sounding and not too invasive de‑esser, plus a gate for the input signal. The Smoothing parameter keys into Humanoid’s FFT‑based processing engine to (I quote) ‘reduce harshness’, but it does so by adding yet another kind of artificial, robotic quality. The Sharpen control is more conventional, applying a presence EQ curve to enliven a drab input signal.

The destructive/creative heart of Humanoid is the Synthesize module. Two separate processes take place here. The first is sophisticated resynthesis, with the module capable of generating a new, synthetic version of your vocal in close to real time. The second, less obvious one is a pitch‑shift that can add a ±1 octave doubling over and above a ±12 semitone transposition to the dry/resynthesized combo. The Pitch section’s scale setting is respected when it does this, which can make for interesting and useful harmonisation possibilities.

In short, there’s a heck of a lot of potential here for huge tonal manipulation, and it’s all invitingly tweakable. The degree of acoustic/synthetic hybridisation is controlled by the big Transform knob, and ranges from no synth contribution at all, through synthetic but still intelligible at 12 o’clock, to complete replacement of the original input and expungement of all consonants, sibilance and fricatives at maximum. A wet/dry mix fader tempers the effect in a different way.

The resynthesis core of Humanoid comes with 64 varied waveforms, but you can also import your own, and then distort them in various ways.

Factory wavetables (64 are provided) range from familiar analogue‑style waves, through metallic and distorted tones, synths and SID chips, to speech/formant, noise‑based and edgy digital timbres. Two parameters, Shape and Stretch, further distort these starting points, increasing harmonic complexity. Further slots are available for user‑imported audio, which can be up to 4096 samples long. Anything longer gets automatically truncated, and in my tests only WAV files (and not MP3 or FLAC, say) were accepted. There’s no support for true wavetable files (in whatever non‑standard format you might have them), but Shape and Stretch distortion of user waveforms is available all the same. Which makes me suspect that the factory waves aren’t genuine multi‑frame wavetables either, but behave as if they are through algorithmic transformation.

I mentioned some vocoder‑like behaviour in the Pitch section, and the Synthesize module can certainly sound something like a vocoder. In fact, though the integration of input and synth that goes on here happens at a much more sophisticated level, it is distinctly more agile, and doesn’t have that typical vocoder squelch. The input signal is the only way of controlling this synth, though, and it has no separate modulation sources, and (in conceptual terms) just one ‘oscillator’.

Moving on, Humanoid’s filter is a potentially resonant high‑pass and low‑pass sandwich with a parametric EQ‑like middle which is rather oddly labelled ‘boost’. A spectrum analyser backdrop lets you easily home in on the harmonic content passing through the plug‑in, and both high‑ and low‑pass filters are capable of dizzyingly steep cuts, paring off individual harmonics if necessary: I’ve rarely seen or heard a filter so ‘surgical’. Bipolar resonance controls also allow for smooth slopes though, as well as more familiar emphasis peaks around the cutoff frequency. The parametric mid band is more conventional, but still very useful for creating scoops and warmth all the way from 41Hz to 12.5kHz.

The Effects section is Humanoid’s last raspberry to convention. Instead of a predicable delay/reverb you get a chorus‑like stereo widener, a Warble effect that modulates the pitch of the resynthesized signal’s upper partials, and a Freeze. That can be clicked or automated to turn Humanoid’s output into a drone or stutter effect, muting the input. The adjacent Buffer parameter controls the length of the loop, with both arbitrary and tempo‑sync’ed timing values.

Humanoid Boogie

Humanoid undoubtedly works best on pre‑existing audio, ie. vocals already recorded in your DAW, as a fair amount of latency is inherent in the design. Buffer size is variable, and at the lowest setting it’s usable in real time, just about, but with a reduction in pitch correction accuracy. Latencies of around 50, 80 and 180 ms for the three available settings were reported in Studio One, which I used for testing, at 48kHz.

In sonic terms, Humanoid’s character is strong, occupying a distinct timbral territory, and it can never be made completely transparent. So it’s unlikely you’ll use it on your next acoustic folk album, but for arresting, creative vocal manipulations it offers exciting options. Alongside the aggressive pitch correction, Synthesize can completely reinvent vocals, transforming them at molecular level if necessary, and the filters can tease out anything from angelic whispers to demonic darkness. It’s worth emphasising here too that Humanoid really does only work on monophonic voices: it leaves polyphonic sources or whole mixes in garbled tatters.

I have a handful of specific criticisms. The steroidal pitch correction, though characterful, is not flexible enough to encompass general problem‑solving: vocals with lots of natural portamento, or that are not well performed or recorded to begin with, can result in an awful lot of adjacent‑note yodelling. I would love to see a way to dial the effect right back. There doesn’t seem to be a way to adjust or tune the correction’s pitch centre currently either. Then generally I found the lack of module bypasses limiting. Not because I’d necessarily want to try and utilise bits of Humanoid independently (which would be wonderful, but I expect impossible because of the underlying processing scheme), but because it could be so helpful and instructive whilst experimenting to temporarily turn off individual bits, without destroying settings.

Humanoid is both inspiring and great fun. It’s quicker and more intuitive to use than any other competing vocal processor I’ve tried and absolutely invites experimentation.

Those criticisms are outweighed by positives, though: above all, Humanoid is both inspiring and great fun. It’s quicker and more intuitive to use than any other competing vocal processor I’ve tried and absolutely invites experimentation. Even from my guttural grumblings it was able to tease out all sorts of great new sounds, ranging from comedic and vintage‑style vocal effects to silky and beautiful synths. There is a ‘sound’, certainly, but Humanoid can easily be different things to different people, with literally dozens of ways to use it. I enjoyed the MIDI harmonisation and filter, and for others it might be all about the tuning and extreme resynthesis. Either way, it’s a plug‑in no otherwordly vocal adventurer should be without.