It looks certain that the arrival of Roland's S770 has signalled new standards in digital sampling. The second part of this in-depth review examines the architecture behind the sound.

LAST MONTH WE discussed the facilities offered by Roland on the S770: compatibility, interfacing, memory expansions and the like. This month we're going to take a look at the sampler's architecture and operation. So without further ado, let's begin with what Roland call Partials.

The Partial Level allows you to combine up to four samples and give each one a separate key-follow pitch scaling, coarse and fine tune amounts, stereo pan and level settings and a velocity window and keyboard crossfade in/out ranges. Bearing in mind that a stereo sample is actually two samples, once again you can apply different settings to each side of the stereo image. As an alternative to fixed pan values, you can set "Key +" and "Key -" - in which case the pan position is determined by the note being played - or Random. The latter is great for getting some spatial movement into the stereo image, and works well for parts that you want accentuated - a synth solo, for instance.

Parameters which affect all four samples alike include output assignment (individual outs 1-6), Partial level and Stereo Mix level (you zero the latter to remove the Partial from the stereo outputs) and coarse and fine tuning. But these parameters are only half the story, because your combined samples can be routed through a digital filter, amplifier and LFO. The Time Variant Filter offers a choice of low-pass, band-pass and high-pass modes, and includes resonance control and a cutoff envelope which provides four times and four levels. Time and level values can be edited as numeric parameters, or they can be edited indirectly by "dragging" points on a graphic envelope display in order to change the shape of the envelope. As with the sample graphic, numeric editing can work best for fine-tuning the parameter values, but in all other respects graphic editing is far superior. You can get very quick results when editing the S770's filter parameters, the only shortcoming being that values aren't updated until you retrigger a note.

Other parameters in the filter section allow you to fine-tune the responsiveness of the cutoff point and the envelope times to the likes of velocity and key-follow; you can also get the cutoff envelope to control the pitch of the sound over time, which can have some interesting results when you're using resonance.

To my mind Roland lead the way with both the quality and the flexibility of their digital filtering. You can get a real warmth out of this filter, but also a really vicious bite, with a cutoff point which can open up and close down faster than a clam on acid. The mere fact that resonance has been included puts Roland ahead of several other manufacturers, and despite what comes across as a "scratchiness" when you use heavy resonance on a sharp filter attack (perhaps because the signal is clipping?), it's a very worthwhile addition, not least as a means of accentuating a fixed frequency. One trick I developed is to assign a rhythm loop to two Partials and leave one of them unfiltered while routing the other through high-pass filtering with the cutoff point jacked up and a touch of resonance to accentuate the high frequency. You can then mix this version in with the filtered version at the Performance level, using it variously to roughen up the sound or add in some "excitement". You can also use resonance in conjunction with a cutoff envelope on a rhythm loop to create some interesting cross-rhythmic effects, or use it to boost the bottom end of a bass sound to devastating effect.

The TVA page follows the same principle of mixing numeric and graphic envelope editing modes, and includes a similar array of parameters for dynamic control of amplitude and the envelope times. In fact, the graphic on each of the pages includes the envelope of the other page as a "background" (non-editable) shape - in a different colour on a colour monitor. Once again, new values don't take effect until you retrigger a note - which can be a trifle annoying if you've set a long envelope release time the only way to cut it short is to exit to the Sound menu). On a more positive note, amplitude fade-out to zero really is very smooth, literally right down to the zero level. You can also set some very long envelope stages - for instance, at 126 (max -1) the release stage doesn't kick in until some two minutes after note release, and then takes another couple of minutes to fade out to zero. Another useful feature of both the TVF and the TVA pages, accessible via the Command menu, is Template. Selecting this calls up a list of ten factory-preset envelope settings; if you then click on one of these envelopes, its settings will be transferred to the Partial envelope. In some cases these might prove adequate as they are, while in other cases they might be good starting points for editing; however, I would like to see the Template principle expanded to include user-programmed envelopes - so that if you come up with a good filter envelope effect, say, you can save it for possible future use on another, perhaps similar, instrument sample.

Finally, the LFO page allows you to choose one of sine, triangle, saw up, saw down, square, random, bend up and bend down waveforms. You can set rate, detune, delay and key-follow values together with key sync on/off and separate mod depth settings for pitch, cutoff and amplitude. Detune is a particularly interesting parameter in that it allows the LFO rate to change as you move up and down the keyboard, with higher parameter values causing greater change; this can help give a more natural vibrato effect on, say, chords played using a string sound.

PATCHES

PARTIALS ACT AS intermediaries between the Sample and Patch levels, and as such all you might want to do with a Partial is assign one Sample to it and give the sample a suitable amplitude envelope. This is because your Patch might simply be a multi-sampled piano or a drumkit spread across the keyboard. Essentially, the Patch level allows you to assign multiple Partials to the keyboard in a multi-split texture. As you might expect, you can do this either by incrementing and decrementing Upper and Lower Note numeric values and clicking on Set or by "dragging" the mouse across the appropriate range of notes on a piano keyboard graphic. In this instance the graphic has another use too, in that you can play notes on the screen keyboard by clicking on them with the mouse. On first encounter the graphic approach to keyboard assignment doesn't seem particularly straightforward, but familiarity breeds content.

Parameters which affect all the Partials within a Patch include octave, coarse and fine tuning, cutoff and velocity-sensitivity offsets, Patch level and Stereo Mix level and output assign. If you set the latter to P, each Partial will conform to its own output assignment - so that, for instance, several drum and percussion Partials can each be routed to different individual outs.

"You can get a real warmth out of the S770's filter, but also a really vicious bite, with a cutoff point which can open and close faster than a clam on acid."

Each Patch can also be assigned a program number for MIDI selection purposes, while, bearing in mind that multiple Patches can be used in a multitimbral sequencing context at the Performance level, you can set Patch Priority on or off per Patch. Priority Patches don't have active notes stolen from them when the polyphony's getting tight. At the Patch level you can also define the response of such Partial parameters as filter cutoff and LFO rate to MIDI pitchbend, aftertouch, modulation and controller messages.

PERFORMANCES

EACH OF THE 32 Parts which make up an S770 Performance can be assigned a MIDI receive channel (1-16), a Patch (1-128), a pan value (L32-R32), an individual audio out (off, 1-6 or Partial), a volume level (0-127), a note range (within 21-108) and upper and lower fade widths (for positional crossfades between Patches).

You can layer Partials at this level simply by assigning the relevant Patches to two or more Parts and assigning those Parts to the same MIDI channel.

If these Partials have been assigned complementary velocity response ranges and curves, you can switch, mix and crossfade between Partials as you can between samples within a Partial. As mentioned earlier, Samples combined within a Partial can't be given separate filter, amplifier and LFO processing. To get round this, you simply assign each Sample to a different Partial and each Partial to a different Patch, then combine them at the Performance level so that, say, four Patches assigned to the same MIDI channel effectively become a "substitute Partial".

But the Performance level's principal application lies with MIDI sequencing. Thus two pages are assigned to MIDI filtering, allowing you to selectively filter out patch changes, pitchbend, sustain pedal, modulation wheel, volume, aftertouch (channel, poly or both) and velocity on each MIDI channel. Additionally you can define a patch map in which Patch program numbers can be mapped to incoming MIDI patch changes.

While Performances allow you to select different instrumental lineups for different songs, they can also be used within songs as a form of automated mixing - snapshot fashion - cutting different instrumental parts in and out and adjusting the level balance between parts. Changing Performances rather than Patches also gets around the problem of having several different Patches responding on the same MIDI channel (as in the "substitute Partial") and wanting to change one or more of them.

Now, you can select Performances from the S770 itself, but a much more interesting method is to select them via MIDI patch changes. You can accomplish this by going to the MIDI Config page in the System mode and setting a Control channel (1-16) and a Control mode (Performance only or Perform/Volume - the latter also allows MIDI patch changes to load Volumes off disk). Now any patch changes received on the control channel will select Performances rather than individual Patches within the Peformance. However, there's a discrepancy in the way that the S770 changes Performances as a result of internal and MIDI selection. From internal selection, active samples continue to play for the duration of their note and their release stage, which is how it should be if you want a smooth transition from one Performance to another. However, MIDI selection cuts all active samples dead as soon as the patch change command is received, which makes life very tricky; in fact, Roland really should change MIDI response to the internal method for it to be of any great use.

"The S770 is a powerful, flexible, extremely easy to use, and for the most part, very well thought out sampler which is sure to win itself a lot of friends."

At the Performance level you also get a very useful MIDI monitor page which allows you to look at the data coming into the S770 (or, equally, see what isn't arriving that should be). You can set the page to monitor all MIDI channels or a specific channel and to monitor or filter out System Real Time messages, while monitoring can be triggered by a specific type of MIDI message.

RESAMPLING

RESAMPLING IS BY no means a trivial aspect of the S770. Mind you, even its simplest function has great value: if you resample a 44.1kHz or 48kHz sample an octave higher than original pitch, again at one of the higher sample rates, and then replay it an octave down to return to original pitch, you've halved the amount of memory needed for the sample. But the really interesting stuff comes when you discover that there's no appreciable difference in sample quality - which suggests that you can effectively double your memory without doubling your memory. The only reason I can think of why this should be so is the S770's Differential Interpolation method of sample playback, which interpolates samples between the actual samples when you pitch a Sample down the keyboard - suggesting in this case that it "regains" top end that has been lost during resampling. Roland make great play of the fact that DI is more intelligent than straightforward linear interpolation, so maybe this is the result. However, this process doesn't work for samples originally sampled at one of the lower sample rates - you get noticeable loss of top end on the octave-higher resample.

The S770's Resampling section allows you to combine two samples into one, which makes it a merge function. But you can also route each sample through its own TVF and TVA in a variety of configurations known as "algorithms". These include not only straightforward sample merging, but also ring modulation of the two samples, so we're talking sound creation here. Additionally, you can set a delay time together with coarse and fine tuning for each sample, which provides even greater flexibility. Incidentally, to resample stereo samples, you do each side of the sample independently and then recombine the two (re)samples using the Set Stereo command.

Resampling is triggered by playing the sample(s) via a MIDI keyboard, which means you can resample samples at different pitches and different velocity responses, and even create new samples consisting of melodies, chord sequences and rhythms. You can also add emphasis when resampling, in order to boost high frequencies. The only inevitable shortcoming of resampling, apart from the fact that you have to loop (re)samples afresh, is that filter and amplitude envelope times and LFO rate become fixed as part of the new sample.

Finally, I must also give brief mention to the excellent manual, which is thoughtfully laid out, for the most part clearly and concisely written, and includes several indices. Whenever I needed an explanation of a particular aspect of the S770, the indices without fail led me directly to the appropriate spot in the manual.

VERDICT

THE S770 IS a powerful, flexible, extremely easy to use and for the most part very well though out sampler which, to use a well-worn cliche, is sure to win itself a lot of friends. The only real disappointments are the lack of sample looping aids and cut 'n' paste-type sample manipulation commands, plus of course the absence of time compression and expansion of samples. The potential to make the S770 as much a synth as a sampler already exists, and in this aim it could be greatly helped by the inclusion of waveform drawing and additive synthesis capabilities (which, after all, aren't unknown on other samplers). I'm sure that some people - particularly those working in conjunction with video and film - would rather like to be able to compile cue lists and trigger them off SMPTE times, but as the S770 can't read SMPTE or MTC I guess this isn't on the cards. Perhaps with the increasing numbers and increasing prominence of hard disk-based recording systems (some of which offer MIDI multitimbral facilities), high-end samplers like the S770 will have to move in this direction if possible. But as a musical instrument the S770 is well impressive.

Prices S770, £4860; M07 Magnetic Optical Disk Unit, £5225; CD5 CD ROM drive, £1340; RC100 Remote Controller, £250; all prices include VAT.

(Contact Details)