|
Use Stereo Samples One of the major problems I had with the AWE32 was that the output from the EMU-8000 was very weak. When digitally recording tracks, the signal never went near 0db. In fact the most it would ever peak at was about -6db, and the average was down at about -10db. I found the problem to be using mono samples for all the sounds. By using stereo samples, the sound becomes much closer to that of the original .wav file the sample was obtained from. The only downside to this is that the polyphony is effectively halved. Still the improvement in sound quality makes it well worthwhile. You might find that you can get away with using stereo samples for the major elements of a tune (drums, bass, lead instruments, etc...) and mono voices for incidental parts (like effects). If you're using a bank editor that is compatible with .sf2 files, you should be able to automatically import 16bit, 44.1Khz, stereo .wav files. If you're using an older one that is only compatible with .sbk files (like Vienna version 1) you can still create stereo samples in your bank. Split the stereo .wav file into two mono ones, then import both, and pan one all the way left, and the other all the way right. Maximise Sample Volume I used to worry about internally overloading the AWE32, by setting the volume of sounds too loud, in both the soundfont bank, and midi sequencer. After learning more of the theory of digital audio, and the structure of the AWE32, I realised that it is quite OK to set the volumes in the soundfont to a maximum. Basically a volume of 127 means that the sample will be played back at exactly the same level as the wave file it was obtained from (although overall amplitude will be reduced slightly as a result of mixing the 32 channels into one 16 bit output). Any value below 127 means that the amplitude will be reduced. Basically you should ensure that all the samples in the banks you create are set at maximum volume. Even if the resulting sound too loud, you can reduce the volume of the track/channel in your MIDI sequencer. Normalise Samples Another necessity for getting the maximum amplitude (and hence best SNR) from your samples is to normalise them before importing into the bank creating software. Most recent .wav file editors include this feature. It basically amplifies the sample, so that the maximum peak in the sample reaches the highest amplitude possible. The diagram below gives you an idea of what normalisation does to the waveform.
This tip is applicable to all samplers, not just the AWE32. As mentioned previously, the digital audio section of the AWE32 is quite noisy, and doesn't record sounds particularly accurately through the analogue line-in (you'll notice that very high and low frequencies are attenuated). To overcome this, try and use samples that have been captured digitally from their source. If you use a lot of material from sample CDs (like myself), use a CD "ripping" program, to extract an exact replica of the sound from the CD. There are heaps of these programs available for Windows 95/98. If you're still using DOS, CDDA is quite a good program (it works with most CDROM drives connected to the Panasonic connector on the AWE32). Click here to download CDDA. Using Different Sample Rates Whilst the EMU-8000 accepts samples in 16 bit, mono, 44.1Khz format by default, it will also accept samples in other formats. You can use samples in 16 bit, mono, 22.05Khz and 11.025Khz formats as well, and such samples will only use 1/2 and 1/4 as much sample memory respectively, as the equivalent 44.1Khz sample. Whilst these sampling rates are no good for high frequency sounds like cymbals, if you are short on sample memory, you could try converting basses and other lower frequency instruments to these sample rates, to save space. The SF2 soundfont standard also allows the use of 8 bit samples, so if you're after some really gritty sounds, and want to save sample memory as well, converting samples to 8 bit might be appropriate. Avoid the OPL3 As mentioned in 'Structure of the AWE32', the AWE32 also contains a Yamaha OPL3 FM synthesiser chip. Essentially this means that the AWE32 has two on-board synthesisers, with each being recognised as a separate MIDI device. However, whilst the EMU-8000 consists of high quality digital sounds, and has the ability for users to load their own samples, the OPL3 is limited to 128 preset General MIDI instruments, and the quality of these instruments is pretty terrible. Additionally the OPL3 chip actually makes a fairly audible 'whining' noise whenever it's operating. Unfortunately because the OPL3 is wired to the S/PDIF output of the AWE32 (along with the EMU-8000) you can't escape this noise by using the digital output, and it is much more apparent through the digital out, than the analogue line-out. However, the way to avoid this problem is to basically make sure that no program on the computer initialises the OPL3 chip whatsoever. So long as it's not initialised, the OPL3 stays silent. Basically make sure that your MIDI sequencer is only outputting MIDI events directly to the EMU-8000 (which is referred to as 'Sound Blaster AWE32 MIDI Synth' in Windows 3.1, or 'Creative Advanced Wave Effects Synthesis for AWE-32' in Windows 95/98), and if you use a program like the Windows MIDI mapper, make sure all events are directed to 'SBAWE32', and not 'SB16'. You can hear the OPL3 noise for yourself, by going into your sequencer and setting the MIDI output to the OPL3 (which might be referred to as 'Voyetra / Sound Blaster SuperSAPI FM Driver'). As soon as you do this, if you listen closely, you should hear a continual, high frequency whining sound. Unfortunately, once it has started, the only way to stop the noise again is to restart Windows, but by preventing any software from initialising the OPL3, you avoid this irritating problem, and get the maximum possible sound quality from your AWE32.
The early versions of the drivers for the AWE32, came with a built in routine that automatically faded-in all the samples. This was great for beginners who inadvertently used samples that didn't start at 0 amplitude, as it stopped the samples from producing clicks and pops. However, for the rest of us it was most frustrating, as it meant many sounds (especially short sounds like percussion) lost all their attack and became thin and frail sounding. I recently found that by using sfstore.dll ver 1.02, this problem was overcome, and the sample was played correctly, with no fade-in. Please note that this file is ONLY FOR WINDOWS 3.x. As far as I know, the sf2 standard did-away with this 'feature', so anyone using windows 95/98, or win 3.x with the latest sf2 drivers should be OK. The best way to test if this applies to you, is to load a sine wave that starts at 32,767 (maximum amplitude) into Vienna (or whatever bank creation software you use). You should be able to hear an audible click at the start of the sample's playback. Click here to download sfstore.dll ver 1.02. Again this file is for WINDOWS 3.X ONLY! Copy it to your /windows folder, and restart Windows for it to take effect. For users of Vienna version 1, please note that after replacing sfstore.dll, you must follow this procedure to avoid samples fading in. For each sample you use in a bank, you must open the envelope deck, and move one of the sliders for the volume envelope (either delay, attack, hold, decay, sustain, or release). I usually just move the delay slider to 1, and then back to 0. If this procedure is not followed, the sample will still be faded in. Unfortunately, to avoid the problem on banks you created with the old version of sfstore.dll, you must open them in vienna, and apply the procedure to every sample in the bank!
As outlined above, the relatively low output of the EMU-8000 necessitates using stereo samples to achieve acceptable volume levels. Unfortunately for AWE32 users with Windows 3.x, this creates another problem (note that this is not applicable to Windows 95/98 users, as the latest drivers for these operating systems seem to have ironed out this problem). It seems that the driver code for Windows 3.x was written such that two samples that are supposed to be triggered at the same time (such as the left and right sides of a stereo pair), do not always sound simultaneously. This is not a problem if the sample is left dry, as because the sound comes from different speakers, you cannot perceive that they are separated. However it does become a problem, when using reverb. This is because the reverb effect on the AWE32 is mixed through both the left and right sides, regardless of which side it emanates from. It results in the two sides of the sample playing slightly out of phase, and causes all sorts of audible inaccuracies, from frequencies being cancelled out, to sounds appearing at random places in the stereo field. These inaccuracies can be identified by listening closely to this example. Fortunately there is a way to work around this. I recently discovered that by only applying the reverb effect to one of the two samples in the stereo pair, the adverse effects of the phasing error are avoided, yet the reverb still sounds much the same (as the effect is sent to both the left and the right channels). I would imagine most AWE32 users would use the reverb effect to some degree on almost every sound, so this technique is essential to avoid these frustrating, and unpredictable problems. High-Pass Filter on the AWE32 One of the AWE32's major shortcomings when compared to other samplers, is it's filtering. The resonant low-pass filter is good, but AWE32 users miss out on the other filter types (high-pass, band-pass, band-reject etc...) that are available on many real samplers. However, using this simple technique you can simulate a high-pass filter on the AWE32. Basically you take the sound that you want to apply the high-pass filter to, and use a .wav file editor to make a second, inverted copy of it (inverting is a function available in most .wav file editors, which makes all positively offset samples negative, and vice versa). Then take both these samples, and import them into an instrument in Vienna (or whatever soundfont creation software you use). In theory these two samples should perfectly cancel each other out, resulting in silence when they play. However what you hear is the original sound, but with the lower frequencies removed, and is very similar to the effect you hear when two stereo speakers are wired out of phase (this is most likely a result of the AWE32 phasing problem outlined above, and the samples not playing at precisely the same time). This cancellation of the low frequencies gives the same effect as a high-pass filter. Now if you apply the low-pass filter to one of the two samples, as the filter takes effect, less and less of the low frequencies are cancelled out. Finally when the low pass filter is completely open, only one of the original samples is playing, and it results in the original (un-filtered) sound. Theoretically you should also be able to apply the low-pass filter to the second sample, but at a different point, and produce a band-pass filter effect. Thanks to Oliver Heusinger for this tip, which was originally posted at Alive! The EMU8000 Level Attenuation and Frequency Response I've recently had the opportunity to perform a series of digital tests, comparing the S/PDIF output of the AWE32 to the .wav files that the sample banks in question were created from. I basically extracted a short clip of audio from a commercial CD, then imported it into Vienna as a stereo sample, and made sure that the volume was set to maximum, and effects (like reverb and chorus) were set to 0. I then played back this sample using a MIDI sequence, with the channel volume, expression and the velocity of the note triggering the sample set to maximum (127), and captured the resulting audio from the S/PDIF output. As mentioned previously, the AWE32 reduces the volume of samples internally, to prevent clipping when playing back multiple voices simultaneously. I actually found that this attenuation is fairly significant. Whereas the maximum digital value reached by the original .wav file was 29428, the same sample played through the AWE32 reached only 13521, a reduction of over 7dB. I then normalised both of the files, and compared them tonally. They actually seemed quite similar, although on closer listening, the sample played through the AWE32 seemed to have lost some of its 'sheen'. On performing frequency analysis on both the files, a distinct roll-off of the high frequencies was present in the AWE32 version beginning at about 14Khz. This might be useful to remember when attempting to master any material created on the AWE32. I also experimented with applying the AWE32's treble controls (available through the AWE control panel with the newer driver sets), and seeing it's effect on the resulting waveform. The AWE32 drivers often boost this treble control by default (which may be an attempt to compensate for the 14Khz roll-off evidenced by the previous experiment). By increasing this treble control, the resulting waveform had a very wide band increase from about 2.5Khz up. Again, if you use the treble control, this information may be useful to you when attempting to master your material. Understanding Envelopes and Looping in Vienna 1 In the past, I rarely used the volume and filter envelopes in Vienna, preferring instead to control filter and volume changes by sending MIDI controller values straight from the sequencer. However having recently experimented with these envelopes, I made some interesting discoveries about how they work. I found that when using a long release time on a sample, the sample would often cut off abruptly shortly after its key was released, instead of fading out gradually and linearly as was expected. At first I assumed this was just another bug in the AWE32 driver, but after some experimentation I discovered the real cause. The AWE32 actually has two different ways of handling looping of samples. In the first case when a key is released (provided the release time is not set to 0, in which case the sample stops playing immediately), the sample continues to play until it reaches the looped region (unless it is already there), and then continually oscillates between the start and end loop points, whilst gradually fading out for the duration of the release time. However in the second case, when the key is released the looping stops and the sample plays through until its end (again whilst gradually fading out for the duration of the release time). Hence (as I discovered), if the length of the remainder of the sample is less than the release time, it can cause the sample to end abruptly before the conclusion of the volume envelope. The choice between these two looping modes is decided by the position of the local loop end point. If the loop end point is left at the default position (ie the end of the sample), then the sample will continue to loop after the key is released (in the former manner described above). However, if the end loop point of the sample is adjusted (even if it is moved to a new position, and then back to the default position), the sample will play through to the end as soon as its key is released, in the latter method described above. At first I found this feature quite annoying, but it can actually be very useful for creating unique effects with samples. For instance by using the second looping type, it is possible to have a sample loop whilst its key is held, and then play a reverb trail when its key is released (thereby allowing the use of deeper and higher quality reverbs than those the EMU-8000 provides, applied directly to the sample through a .wav editor). An example of looping such a sample is shown in the figure below. It would even be possibly to create a sample which played one sound when its key was pressed, and a completely different one when its key was released. NOTE that the techniques described above only work in Vienna version 1.X. Vienna 2 introduced a check box in the looping options marked 'Release after loop region', which enables manual selection between the different looping modes.
Whilst many of the problems I have found with the AWE32 have straightforward resolutions, there are still two outstanding issues which I have found no way to work around. The first concerns problems with panning. If you use a stereo sample (or two mono ones panned hard left, and hard right respectively), when using MIDI panning (controller 10), values of 0 and 127 will not produce sound that is panned completely to the left or right. The diagram below depicts a synth-lead line that has been panned all the way to the left (controller 10 set at 0). Despite this, it's obvious that there's a significant amount of sound coming from the right channel (shown as the bottom waveform). Basically, with stereo samples, MIDI controller 10 will pan the sound, but not completely to the left or right. I have tried doing the same thing under Windows 95, and it produces the same result, indicating a limitation of the AWE32 hardware, rather than a software bug. There are two ways to work around this, although I consider neither to be completely acceptable. One suggestion is just to use a single mono sample, panned to the centre. The major problem with this is that the sound produced will be weak by comparison to stereo sounds in the same mix. The other option is to use two identical mono samples, both panned to the centre. Although the sound will pan correctly using this method, because the two samples are played slightly out of phase (as described in "The AWE32 Phasing Problem" above), it can result in audible inaccuracies in the sound (like frequencies being cancelled out).
|
