Yamaha SREV1

Yamaha' s cutting-edge reverberation processor uses convolution techniques to sample the ambience of real acoustic environments, allowing extremely accurate simulations.

Photo: Mike CameronThe science of artificial reverberation has advanced considerably in the last decade, and particularly in the last few years with the development of convolution technology. This enables the acoustic signature of real reverberant spaces to be recreated with astounding precision. Although the maths behind convolution has been known for a while, the vast amount of DSP required restricted it to off-line uses until very recently.

All that changed, though, when Sony launched their DRE S777 convolving reverb machine a couple of years ago. Yamaha have also been developing such processing for many years, and have now launched a competitor to this system with their SREV1.

For a detailed explanation of convolution, take a look at the DRE S777 preview in SOS June 1999. Basically, a specially recorded sample (containing the impulse response) of a real acoustic reverberation pattern is imposed on ('convolved with') another audio signal. The result is a reverberant signal with many of the sonic characteristics of the original acoustic space.

True, for typical pop-music applications, the advantages of improved reverberation fidelity will probably be lost in the density of the mix. Besides, realism is rarely a requirement in this kind of music. However, in situations where the reverberant space is a more obvious and crucial element of the sound balance — such as in acoustic music of all types, or in cinema and TV soundtracks — the increased realism provided by convolution is very worthwhile.

To many, though, the biggest advantage of convolution is that the user can sample their choice of acoustic space and then recreate its acoustic signature to order. This can potentially prove extremely valuable for classical music and film sound applications.

Yamaha's Convolution Technology

Yamaha have a long tradition of designing and producing their own bespoke integrated circuits where conventional off-the-shelf components do not meet the company's specific requirements. It therefore comes as no surprise to learn that the SREV1 employs no fewer than 32 DSPs designed specifically to perform the many calculations required for real-time convolving reverberation — up to 520,000 steps per sample! These provide 24-bit audio I/O with 32-bit internal processing (44-bit in the EQ stages) and support 44.1kHz or 48kHz sample rates.

The machine has three operating modes: high-resolution stereo, four-channel surround, and a lower-resolution dual-stereo mode which provides two fully independent stereo processing paths. As standard, the machine can provide up to 5.46 seconds of convolution processing per channel in stereo, and 2.73 seconds in four-channel surround or dual stereo modes. However, an optional DSP extension card (installed in the review machine) adds a further 32 convolution chips to double the processing capabilities and increase the real-time convolution up to 10.92 seconds in stereo and 5.46 in surround or dual-stereo modes.

It is worth noting that these 'convolution times' are related to, but different from, the 'reverb time'. The normal specification of a reverb (decay) time is the time taken for the amplitude of the reverberation to decay to 60dB below its original peak. However, with 24-bit audio resolution the decaying reverberant tail must be generated (by convolutions) all the way down to -120dB and beyond. Therefore the convolution times quoted translate roughly into maximum reverb decay times of about half as much — the basic machine in stereo mode will cater for reverb decays up to about 2 seconds, falling to just over a second in the four-channel and dual-stereo modes.

Clearly, if you are using the sampled acoustic reverberation of a real physical space, the comprehensive parameter editing facilities found in typical artificial reverberation machines are meaningless. Instead, the SREV1 provides facilities to change only the initial delay, and the overall decay time of the reverberation. Added to that, the user can adjust the dry/wet balance and adjust pre- and post-reverb equalisation with four-band EQ stages on each channel. Although this sounds rather restrictive compared to conventional reverb processors, you have to remember that the whole raison d'êtreof this machine is to reproduce the real reverberation characteristics of specific locations, not to enable the user to construct artificial reverberation. Nevertheless, the available parameters are sufficient to tailor the sound of the reverberation to suit the material, providing much the same kind of control that might be obtained by repositioning 'space' mics in a real acoustic environment, for example.

Hardware Facilities

Yamaha's SREV1 is a 3U black rackmount box with no front panel controls and only a few LEDs, and it is controlled from a large, but attractive remote panel, complete with four motorised faders. The resemblance to the Lexicon LARC is surely no accident, although its rather larger footprint may make it less easy to accommodate on a console surface. One controller can operate up to four SREV1 units, if required. The remote control is normally powered through its interface cable from the main processor unit, although an external power supply can also be used if necessary.

The main processing unit carries a CD-ROM tray and PCMCIA memory card slot on the front panel (both Windows compatible), along with the power switch. Status LEDs indicate the presence and clipping of input and output signals, and the sampling rate. The rear panel is equipped with two AES-EBU inputs and outputs as standard (providing four channels of audio I/O) plus two mini YGDAI slots to allow customer-specific interfacing (using Yamaha or Apogee cards). The review model was fitted with Yamaha's MY4AD and MY4DA cards, providing four analogue inputs and outputs. There is also a pair of MIDI sockets (In and Out), plus two mini DIN serial ports, a word clock input BNC and a D-Sub connector to interface with the remote controller. The serial ports are intended for interlinking multiple SREV1s.

With so much number crunching going on, heat is an inherent by-product, so the machine is force-cooled by two fans mounted at the rear. These suck air through the machine from a grille in the front panel. The expelled air is noticeably warm, but the rest of the machine stays remarkably cool. However, these fans are relatively noisy, so it would be worth putting the unit in a separate machine room.

With so much number crunching going on, heat is an inherent by-product, so the machine is force-cooled by two fans mounted at the rear.The remote control is supplied with a 20m cable fitted with D-Sub connectors to carry both control data and power. The control data can also be run over greater distances using standard data cabling, but then a local 12V DC supply must be used to power the controller via a coaxial socket on its rear panel. Also on the rear panel are brightness and contrast controls for the LCD screen.

The front panel has four motorised faders along with a data wheel, cursor, Enter and Inc/Dec buttons. There are also five further buttons to access programs, main and fine parameters, utility functions and a bypass mode. Eight LEDs indicate clipping on each input and output, and a large, backlit 320 x 240-pixel LCD screen shows the current program information, configuration menus, and the labels for the software functions allocated to the four faders.

Hooking It Up

Connecting the SREV1 to a digital console is perfectly straightforward using the AES-EBU inputs and outputs. A suitable aux send is derived from the mixer and the output from the SREV1 returned to a convenient digital input. The SREV1 is then configured to use the AES connection as its input signal source via an input patch menu page, and also to use the AES input as its word clock source. The latter ensures that the reverb return is synchronous with the word clock of the mixer.

If the optional analogue interfaces are being used, the appropriate YGDAI inputs are selected and the SREV1 is configured to operate from its internal 48kHz word clock. The input source and word clock selections are made in one of the utility menu pages, along with various metering modes, input and output level adjustments, and a bypass mode selection. There are two bypass modes called Direct Out and Reverb Mute, but there is no reverb input mute facility. The Direct Out mode bypasses all digital processing between the input level and output level stages, whereas the Reverb Mute mode disables only the convolving and EQ stages, retaining the reverb balance and level parameters. In my experience, a more useful bypass mode is to mute only the input signal such that the reverb tail continues to die away naturally when the source is muted. Although this function can often be performed by muting the console's aux send, having it on the remote controller makes life easier.

Programs, Projects & Impulse Responses

The creation of a reverb program requires the appropriate impulse response data taken from the desired acoustic environment. Pre-recorded data supplied by Yamaha is stored in an internal memory at the factory, as well as being included on a CD-ROM shipped with the machine. There are some 79 different reverberant environments available on this first disc, although others are available (see the 'Additional Programs' box). Users can also sample their own acoustic spaces with relatively simple facilities and the supplied software (see 'Sampling For Yourself' box).

The impulse response data of the acoustic spaces is organised in stereo, four-channel surround, and dual-stereo configurations for the corresponding programs to use — although only those pertaining to the current operating mode are shown in the menu selection lists. The programs and the drives or memory in which they are stored are managed and accessed through pages in appropriate menus.

A selection of 79 sampled acoustics are shipped with the SREV1, including some derived from highly-regarded performance spaces.The factory preset programs (constructed from the impulse response data) are stored in the machine's internal memory, as already mentioned. The program data includes information about the adjustable parameters such as reverb time, initial delay, EQ, wet/dry balance, levels and so on. User programs derived from the original factory presets can be stored in this memory too, or on a removable memory card, or in internal Quick Memories. The difference between Quick Memories and the internal or card memory is that programs can be loaded instantaneously from the Quick Memories, whereas the internal and card memories are considerably slower to load. There are six Quick Memories available in the stereo and four-channel modes, and 12 in the dual-stereo mode.

Projects contain all the Quick Memory programs as well as the current program, allowing all settings for a specific project to be collated and saved to either the internal memory or a card, and recalled en masse when necessary. The number of projects or programs that can be stored in the internal memory or card depends on the capacity of the memory and the size of the associated impulse response data. Strangely, the physical configuration of the machine is not stored as part of the Project data — setup data like the reverb mode (two or four channels), I/O selections and levels, word clock source and MIDI settings, for example.

Using The Yamaha SREV1

The SREV1 is simplicity itself to configure and use. Selecting and loading programs is intuitive and, although the structure of projects, libraries and Quick Memories is confusing at first, the organisation becomes clear enough after a little poking around.

The use of faders on the remote allows fast and accurate setting of the various parameters, provided a little care is taken to observe the LCD screen to make sure you are about to adjust the intended aspect of the current program. In practice, I found only being able to adjust the reverb decay and pre-delay times of the active program quite adequate to fine-tune it for a particular application. The reverberation is so natural and usable in the first place that only a small amount of 'damping' and distance needs to be tweaked to suit the source material. It is slightly disconcerting, though, that the reverb output mutes momentarily as the reverb decay time parameters are adjusted — even if it is preferable to the nasty glitching that would otherwise become audible as the convolution parameters are changed.

Although there are very few adjustment parameters for the reverb processing, onboard four-band parametric equalisation on each channel, both before and after the reverb processing, allows a great deal of sonic fine-tuning.The built-in four-band equalisers both before and after the reverb processing are typical Yamaha designs, with all the usual facilities you would expect, including switchable filter types plus gain, frequency and bandwidth controls. I was slightly surprised that tailoring the input EQ slightly had a dramatic affect on the perceived quality and nature of the reverb — probably more so than using EQ on the output side, in fact. In concert with the adjustable decay time, this proved more than sufficient to adjust the reverb sound for almost any situation.

Since the ability to tweak an SREV1 program is very much more limited than it would be with, say, a Lexicon artificial reverb algorithm, it is far more important to select the best starting point in the first place, and that means taking the trouble to listen carefully to the 79 different programs on offer and really become familiar with their characteristics. There are plenty of well-recorded samples of fine European, American and Japanese acoustic spaces supplied with the machine, and I was able to find near-ideal reverberations to match a wide variety of mainly orchestral and choral material that I happened to be working with during the review period. Unfortunately, though, there seem to be relatively few small-room acoustics supplied with the SREV1 — the majority of programs are large or moderately-sized concert halls, churches, studios and stages, plus a few very good plates. Since it is small rooms which are usually the hardest to synthesise convincingly with conventional digital reverb units, this seems a wasted opportunity. Of course, the user could sample their own small rooms if required (although the opportunity did not arise for me), and all that is required is a fairly modest PA, some mics and preamps, the SREV1 and the time. Nevertheless, I would recommend that Yamaha sampled a wider range of spaces and didn't just concentrate on the more expansive rooms. Presumably, additional reverb programs will be made available as the user base for the SREV1 grows anyway, and it is easy to exchange program data with other SREV1 users over the Internet as standard WAV and AIFF audio files, if you want to.

Yamaha's SREV1 is designed from a different standpoint to Sony's DRE S777, and will probably appeal to different users. The SREV1's large and noisy hardware is clearly designed to be located in a machine room and controlled through its elegant remote control. It also offers slightly more in the way of tweakability thanks to the two EQ processors. However, I would strongly recommend trying both units if you are considering investing in a convolving reverb machine, to see which suits your particular application best. Both sound superb and demonstrate the clear advantage of this approach to reverb in the right circumstances. Despite the falling cost of DSP, the processing requirements of sampling reverbs are so vast that this technology is very expensive in the UK compared to more conventional methods, and is therefore hard to justify for all applications. Nevertheless, where precision and naturalness are important, this is the way to go, and Yamaha have a strong product in the SREV1.