The Voice One offers the second generation of TC-Helicon's pioneering modelling technology, along with powerful tools for manipulating and correcting pitch.
TC-Helicon's first major products, the Voiceprism Plus and the Voicecraft processing card, have already built the company a name for innovative voice processing based on sophisticated modelling techniques. Although sharing some basic functionality with the Voiceprism Plus, the company's latest processor, the Voice One, is a markedly different tool intended to fulfil a rather different role. Whereas the Voiceprism Plus contained a mic preamp and offered various conventional effects and reverbs, the Voice One is a pure voice manipulation tool. It provides only the most sophisticated voice modelling and pitch-correction facilities possible, and is a serious, dedicated studio tool, rather than a general-purpose multi-effects unit.
The Voice One combines pitch-shifting and physical modelling processes into a single box, but the code hasn't just been lifted wholesale from their existing products. For example, a new 'vocal-trained' pitch-recognition algorithm makes the pitch-correction particularly impressive, and the second generation of their voice modelling has been redesigned for higher fidelity. Furthermore, where vocal quality is paramount, there is a dedicated Pure Shift mode which applies all the DSP resources to the pitch-shifting algorithms. This provides better quality and lower latency, but disables those voice modelling functions not related to pitch-shifting. There is also a new Flextime process, which is designed to provide more natural pitch and phrase control, which can be used to add human-like variations in timing and pitch when generating a double-tracking effect or harmony part.
As you might expect, pretty much every parameter can be controlled in real time over MIDI — see the 'MIDI Control' box for details. The range of parameters provided to fine-tune the algorithms is mind-bending, and it is possible to create some stunningly realistic effects with care. On the other hand, it is very easy to create complete nonsense — although some of the results I created by accident proved entertainingly useful in their own right. Fortunately, the machine is supplied with 100 factory presets which act as excellent starting points for most requirements, and there are pre-programmed Styles for each of the voice modelling functions to simplify the parameter configuration whilst maximising the creative possibilities.
The fact that the Voice One has been designed for serious studio use is hinted at in the physical styling, since it looks like a conventional effects processor. There are also no mic preamp facilities, and only a 'wet' output is provided, which means that it needs to be interfaced through a mixer. Whereas the Voiceprisms were 2U boxes, the new Voice One is a slim 1U rackmounting unit, with a similar layout of rotary controls, display panel, and mode-selection buttons as found on TC Electronic's M*One and D*Two products. The basic operation is also very similar to these units, both in configuration and general usage.
Although the dual-channel analogue and digital audio interfacing on the rear panel of the Voice One would suggest otherwise, there is actually only one active channel. The provision of the second channel's I/O probably has more to do with maintaining commonality of internal hardware with the TC Electronic processors than any operational considerations. Nevertheless, TC-Helicon have used the facility constructively, allowing the user to select the input from either one of the two channels, and to output the processed signal and a delay-compensated direct output simultaneously. This feature can be used to ensure that the original source and a created harmony part, for example, remain in time with each other.
The analogue inputs are electronically balanced on XLRs and accept either professional (+4dBu) or consumer (-10dBV) line levels, selected through a configuration menu. In addition, a front-panel rotary gain control allows fine adjustment over a 24dB range. A pair of bar-graph meters to the left of the display window show the input and output levels with a 40dB range.
Like the analogue input, the output level is configurable too, with four operating ranges. These provide peak levels of +20, +14, +8, or +2dBu — the last being compatible with -10dBV systems. Again, the precise level can be adjusted over a 24dB range through the second front-panel rotary control. The unit also has a configurable high-pass filter to remove low-frequency information which would otherwise degrade the performance of the pitch-recognition and voice processing algorithms. The turnover frequency can be set manually or automatically, the latter adapting to the signal content.
The A-D and D-A converters are the usual oversampling delta-sigma types, which impose roughly 0.7ms of conversion delay each (the precise figure is dependent on the sample rate, of course). Thus, using analogue signals in and out of the machine will impose about 1.4ms of delay, although this is swamped by the inherent signal processing latency of the various modelling effects. Whereas the processing delay can be accommodated fairly easily in a studio situation, it can become critical in live applications. Fortunately TC-Helicon have thoughtfully provided two alternative operating modes with different latencies. The downside, naturally enough, is that the quality of the processing tends to suffer with lower latency.
The Voice One's digital I/O is via unbalanced phono sockets, but the data can be switched between AES-EBU or S/PDIF modes, as required. There are three clocking options: internal 44.1kHz and 48kHz, or the rate at the external digital input. The digital input can also be adjusted in level via a menu, and the output resolution dithered to 24, 20, 16, or 8 bits, allowing optimised connection with a wide range of systems.
The Voice One is supplied with 100 factory presets (and 50 user memories), starting with voice modelling treatments such as Teen, Older and Whisper — which modify the source voice according to the description, the last being quite eerie! The preset list carries on through single effects such as the weird Flextime Extreme and various double-tracking effects, basic harmony generation, and pitch-shifted voice effects (such as a very Monroe-esque model).
There follow various pitch-correction modes, including some bizarre ones such as Invert, Compress, and Expand, all of which mess up the source pitching in pretty dramatic ways. A collection of MIDI-controlled presets are followed by some more realistic pitch-correction modes and then various voice modelling functions including the scarily impressive Whiskey and Marlboro presets, and a Stretch Neck patch which sounds exactly like its name suggests. A small collection of special effects presets includes a Can't Sing function which is far too reminiscent of studio sessions best forgotten, but I dare say it will be used to abuse singers up and down the land to everyone's great mirth...
Although there is the usual array of entertaining but 'never to be used in earnest' presets here, there are also a great many which provide extremely useful functions, or act as good starting points to fine-tune something specific. For example, I found the Tight Double and Scoop Double presets gave very effective and realistic double-tracking effects, although there is an appreciable delay when operating in standard mode. (I didn't measure it exactly, but would estimate it to be maybe 100ms or so). Setting the machine to low-latency mode removed virtually all of the delay and made live performance voice processing quite practical. The quality and realism of the double-tracked effect was certainly not as good as in the standard mode, but in the context of a one-off live performance it was perfectly acceptable and far better than any other technique I know of.
Pretty much every function of the Voice One can be controlled over MIDI. Besides the expected preset recalls, there are also dedicated MIDI channel controllers for many of the parameters, including pitch-shift and pitch-correction amounts, scales and base keys, and also breath, growl, and vibrato values. If that is not enough remote control, every single parameter of the entire machine can be controlled through SysEx commands. The handbook allocates no fewer than six pages just to the MIDI functionality, which is impressively comprehensive.
The Voice One is configured through two buttons labelled I/O and MIDI/Util, along with Recall and Store buttons to access and save preset memories. Two further buttons on the right-hand side, labelled Enter and Bypass, are self-explanatory — however, a footswitch socket on the rear-panel also allows remote bypassing of the processing, allowing a harmony or double-tracking effect to be punched in and out during live performance, for example.
Each voice-modelling section has a dedicated function button (Inflection, Spectral, Growl, Vibrato, Resonance and Breath), along with two further buttons for pitch processing (Shift and Correct). Each of these has an LED in the corner, with the function being active if the light is on. The modelling blocks accessed via these buttons allow various facets of the human voice to be manipulated to change its character — or indeed, create a completely different character — with remarkable realism if treated carefully. Double-tapping these buttons allows editing of the relevant algorithm's parameters. A single-line alphanumeric read-out on the display identifies the current parameter and up/down arrows indicate where the others are located. Up/Down buttons allow the different parameters to be selected, and the adjacent rotary encoder changes their values.
The display panel is a multi-coloured affair with a pair of green bar-graph meters to the left showing input and output levels (and revealing the processing latency quite clearly!). The upper section of the display is divided into two, with the left section representing pitch-shifting and correction and the right relating to the voice modelling.
The pitch display provides two horizontal bar graphs, the top one showing intonation in cents above or below the nearest recognised pitch, with the second bar graph showing the amount of pitch-correction being applied. In addition, the recognised source pitch is also shown by circles dotting about on a one-octave keyboard display at the bottom of the panel, and by a note character readout. The pitch-correction scale and base key are also indicated by both character read-outs and solid dots on the keyboard display. The voice modelling window comprises six short vertical bar graphs showing the amount of processing applied by each of the six VM algorithms. A line of illuminated flags underneath show the current status of the machine, such as whether a factory or user preset is in use, whether it has been edited, if MIDI data is being received, and whether the machine has recognised the input signal.
As you can imagine, the voice modelling tools and pitch-shifting algorithms provided in the Voice One are about as complex and state-of-the-art as you can get, so to cut down on the complexity as far as possible and make this machine reasonably easy to use, TC-Helicon have provided a small collection of variable parameters for each algorithm, supplemented by a range of Styles. The latter effectively recall pre-programmed parameter sets for specific musical applications, neatly avoiding the user having to fine-tune several dozen parameters. The result is a surprisingly wide range of effects, with simple user control.
For example, the Breath voice modelling section is equipped with user parameters for Amount, Harmonics, and Style — the last offering 29 variations including such options as: Light, Airy, Intimate, Dirty, Raspy, Phlegmy, and Whispers — to name just a few. Each voice modelling section has its own set of Styles, allowing any kind of combination you can contemplate. There are 28 Spectral options, 36 Vibrato, 12 Inflections, 20 Growls, and 22 Resonance Styles — that's a lot of creative possibilities. Okay, so some are more effective and usable than others and, as with all signal processing, the more subtle the approach the more believable the end result. Nevertheless, the creative possibilities here are awesome.
Similarly, the pitch-correction algorithms are equipped with comprehensive user parameters, including 49 musical scales plus a custom setting option. The built-in scales span all the usual western modes including relative rarities like Mixolydian and Locrian, before heading off into the distinctly obscure such as Neapolitan Minor and Hungarian Gypsy. The world trip continues with scales that, I must admit, mean little to me, such as Hirajoshi, Kokin, Ritusen, Pelog, Arabian and Persian, to name just a few. Once a scale and base key have been declared (either through static manual settings or via real-time MIDI), the user can tweak the correction algorithm's parameters to optimise the tuning error window (how far out of tune the source has to be before it is corrected), attack time (how quickly it is brought back), and amount (the proportion by which the tuning error is corrected).
The pitch-shifting algorithms are even simpler to set up. The amount of pitch-shift can be determined up to ±2 octaves, through one of four basic modes. A chromatic mode allows the shift to be defined in terms of cents (±2400), while a second option allows the selection of any musical interval up to an octave in one of three different major and minor scales. Pitch-shifting can also be based on the selected pitch-correction scale, or a custom scale whereby any incoming note can be mapped to any new output pitch.
If using the Pure Shift mode — where the full DSP power is allocated purely to pitch-shifting algorithms, disabling the voice modelling functions — an extra parameter is made available to adjust the formant character. Positive values have the effect of extending the modelled vocal tract for a deeper, more masculine sound, whereas negative values make for a shorter tract, simulating a younger voice. Used sparingly, this is a remarkably powerful tool. It is also possible to determine how much formant correction is applied for a given pitch-shift amount, since small shifts benefit from little formant tweaking whereas large shifts require much heavier correction to avoid the classic 'chipmunk' effect. An automatic mode varies the amount of formant correction dynamically for the most natural results.
TC-Helicon have been very keen to point out that the Voice One uses second-generation voice modelling algorithms. One of the criticisms levelled at the Voiceprism Plus in SOS was that even with everything set to zero, just passing a voice through the VM processor resulted in a distinct change to the source quality. I'm therefore very pleased to report that engaging all the VM functions on the Voice One, but with their values set to zero, had negligible effect on the source voice — I certainly couldn't tell any significant difference while switching the bypass in and out (apart from a strange phasey effect which occurs briefly as the bypass is activated). This is an important improvement over the previous generation of VM algorithms, as it means you can now introduce the required degree of voice modelling without affecting the voice in any other unintended way. A worthwhile upgrade.
The development of voice modelling has progressed enormously in a relatively short time, and the Voice One really does represent the current state of the art. Although the technology hasn't quite reached the stage of turning 'the bloke down the chip shop' into Elvis, it can certainly make a jolly fine stab at the reverse! Broadcasters will certainly want to investigate this machine for disguising voices in sensitive interviews. Indeed, a few tests quickly proved that even close relations could be fooled, yet without damaging the intelligibility of speech in any significant way. A very impressive result.
This machine is also very powerful when it comes to changing the characteristics of a voice to suit different song styles — for example to add a gritty quality to a 'choir boy' voice, or to add a little weight and breathiness for a smoochy ballad. Okay, so the purists (and I guess that includes me) would argue that it is far better to match appropriate songs to the singer's particular talents in the first place, but I appreciate that many will find a tool of this kind very useful indeed. Like any effect, though, these voice modelling tools really have to be used sparingly if they are not to stick out. Increasing amounts of each process quickly produce mechanical results and the illusion is destroyed completely.
However, the pitch-shifting and pitch-correction tools are superb. Although the raison d'être of the Voice One is its voice modelling capabilities, the fact that it also incorporates pitch-correction and formant-corrected pitch-shifting will act as a significant incentive for would-be purchasers. The industry standard for pitch-correction has been defined by Antares, of course, although I feel that perhaps TC-Helicon are taking up the challenge. Indeed, although I was not in a position to compare the Antares and TC-Helicon pitch-correction capabilities directly, my gut feeling is that the latter imposed a slightly more natural, less mechanical character while correcting pitch.
I also found the double-tracking and harmony functions, in particular the Flextime feature, to be very good indeed. The random timing variations really helped to make the doubled part sound convincingly real, especially when combined with small pitch variations. By using the pitch-shifting mode as well — especially if controlled by MIDI to play the appropriate harmony line — the results were stunningly realistic. In one test I shifted Suzanne Vega's voice to create an incredibly believable male countermelody part for 'Tom's Diner'.
The processing latency is significant, though, when working at the highest resolution. Disabling the voice modelling functions to benefit from the Pure Shift mode helps considerably, and I found I tended to work in this mode more than any other. In the studio, the ease with which tracks can be re-timed in DAWs means that the latency is not really a problem anyway. However, for live stage use the low-latency mode is essential. Although there is a noticeable reduction in processing quality if auditioned critically, I don't think anyone would notice in a typical live stage setting — particularly if the unit were only used occasionally to add dramatic effect. The Voice One is certainly an impressive tool, and one which redefines the genres of voice modelling and pitch-shifting. Having said that, this is still a way off being a perfected technology, and misuse or overuse quickly becomes audible. Used with care, though, the results are stunning.
To develop the technology which drives the Voice One, TC-Helicon have obviously had to do their research, and if you're interested in this aspect of the product then you'll be pleased to know that some of this research is published in the Knowledge Base zone of their web site. The document to download is the one titled 'Pitch-shifting & Voice Transformation', by Patrick Bastien.