MIDI Matters (Part 4)
Sample Editing Via Midi
Why is it there are so many computer-based sample editing packages being released? Perhaps it has something to do with the fact that they make working with a sampler considerably easier. Jay Chapman ponders their worth.
This month's article follows on from the April issue where I looked at how a computer, acting as an intermediary between a MIDI-equipped synthesizer and its owner, might help with the complex and often tedious task of parameter editing: this month I look at a similar set-up where a sampler rather than a synthesizer sits centre stage.
The business of altering parameters via MIDI has not changed so I can let you skim over April's article to refresh yourself with the basic ideas that apply. The use of graphics, colour, and so on - perhaps all driven by a mouse - is also just as important as before, as is the idea of grouping sets of parameters into screen pages to emphasize context and help the user mentally model the device he seeks to control.
What is very different when sampling rather than synthesizing is that you can have at your fingertips the data which, effectively, is the sound that will eventually issue forth! So, we can play with the sample data itself, in fact we can often get right down to an individual sample datum (perhaps an 8-, 12- or, if you are really lucky, 16-bit number), and modify it at will.
In the case of the synthesizer, we were regulating control values which dictated how the synthesizer components were configured and how much each contributed to the creation of the sound being output. I grant you that this is something of a simplification, but consider a typical 'subtractive' synthesizer for a moment. We select the oscillator waveform from a set of maybe four or five available waveforms and that's the last time any sound creation is done. From that point on, we are modifying and often removing something from the sound. The various control values vary the quality and quantity of modification performed by each device in the signal path.
For the sampler, on the other hand, it is often the case that the sound is not altered radically once it has been produced (often just some low pass filtering and envelope shaping). We can modify the potential sound quite considerably, however, by modifying the numbers that represent it in the memory of the sampler.
This is a nice example of two types of device which compliment each other well. The ability to record and replay samples is provided by the specialised hardware in the dedicated sampler. Many, if not most, 8- and 16-bit microcomputer systems would not be able to cope with medium to high quality sampling even with the addition of suitable analogue-to-digital and digital-to-analogue peripheral equipment. They just don't have enough muscle to handle sampling along with any other useful work.
In general, the specialised sampler does not have suitable facilities to improve the human/user interface and so can be greatly enhanced by the addition of even an 8-bit microcomputer dedicated to this task. This is not to say that manufacturers are not aware of the problems facing the user. The Roland S-50 with its RGB interface for a colour monitor and the Casio FZ-1 with its 64 x 96 dot-matrix LCD bear witness to the fact that manufacturers are trying to squeeze in some user-friendliness without offending your wallet too much!
To be fair, the current generation of extremely powerful microcomputer systems based on the Motorola 68020 and Intel 80386 microprocessor chips shouldn't have any problems dealing with sampling and the human/computer interface at the same time - the Commodore Amiga has already been setting the scene. That's another story though, so you will have to keep buying Sound On Sound to find out what the future holds in store. You'd better start saving your pennies too... the nice toys that are coming will be very expensive at first!
We can assume that with sample editing software you are going to get all the good stuff that software editors for any synthesizer should give you. In particular, though, you would expect some facilities that are related directly to sample editing and the use of samplers in performance.
WAVEFORM DISPLAY: Perhaps the first facility I would expect is that of graphically displaying the waveform corresponding to the sample on the microcomputer's screen. There will probably be a number of different ways of doing this involving varying degrees of magnification. If you are examining a one-second sample consisting of in excess of 40,000 samples, you can't just put up one pixel (ie. dot on the screen) showing the size of each sample - that would require something like 80 screens, side by side!
It is useful to be able to see several cycles of the sample waveform at different points throughout its duration. There may well be a pseudo 3-D mode where successive sections of the waveform are displayed slightly above and behind each other (a la Fairlight CMI) to give a sort of overview of the whole sample - see Figure 1 which shows a display from the Sound Lab visual editing system for the Ensoniq Mirage sampler. In either case it may well be that only every 8th, 16th or 32nd sample datum is actually displayed so that a useful length of time is represented by the width of the screen display.
Once you can display parts of the waveform on the screen to different levels of accuracy, you can move on to editing the waveform (see Figure 2). One technique is to 'draw' a new section of a waveform by moving the mouse pointer, or sometimes a light-pen, around the screen display. Imagine that you have merged the front end of one sample with the back end of another ('PianoString' is the obvious example) and have created a noise glitch where the two waveforms join. All you do is display the joint on-screen and draw in a nice new transition curve replacing any vertical lines that are sending your speaker cones into outer space!
In Figure 3 you can see how a Loop Splice point is selected using the Sound Lab software by viewing the waveform with sufficient detail to make out a cycle and then moving a vertical cursor line to select a zero crossing point. In this way, loop points can be set up in seconds rather than minutes, which might be important if studio time is ticking away...
By viewing sample section end points before replicating sections to form complete samples, it is possible to manipulate the sample data to avoid both noise spikes and changes of phase at the joins. It is also possible to scale and compress data and to automatically interpolate between the end of one waveform section and the start of another over a period of time.
CUT AND PASTE: Wall-papering in Sound On Sound? Not quite! It might be fun, and might also produce some interesting sounds, to use cut and paste facilities to move sections of various waveforms around to create new waveforms. 'Cut and paste' says exactly what you do! Imagine drawing sections of waveforms on a piece of paper, cutting them out, fitting them together in different sequences and then pasting the final version together as a new waveform. Using the computer's mouse you direct the software to do exactly the same work with the displays on the screen - but with a lot less mess and a great deal more ease.
For example, it can be a fairly simple matter to work out where the phonemes start and end in a vocal sample of the word 'n-ineteen' by doing some experimental cutting and listening to the results. If you view the sample on a low magnification setting, you can then mark off the 'n' and repeat (ie. copy) it a few times. Insert the repeats before the leftover '-ineteen', and there you have it - instant fame and fortune!
If you have managed to find software that will also act as a sample librarian, you will soon get used to the idea of pulling up several samples on-screen, cutting them up a bit, and mixing and merging them to produce new sounds. In many cases, the librarian will also handle multi-samples for you.
MULTI-SAMPLES: If you carefully sample your parents' Steinway Grand on middle C only, and then try to replay the sample over the whole keyboard range, you are in for some disappointment. Having read all the Sound On Sound back issue articles on sampling, you go back down to the drawing room intent on sampling at least one note per octave. You then have to organise the right samples to be played by the right keys on the sampler's controlling keyboard.
In this case it might be useful to assign different samples to their logical positions on the sampler's keyboard by using the mouse to select the range of notes on a mimic keyboard display. No key numbers have to be worked out in hexadecimal, you simply point at the first key and then drag the mouse along until the cursor points at the last key affected. Each key changes colour on the display as you select it. Any keys that haven't changed colour by the time you have used up all your samples, you worry about!
Another obvious utility is to allow samples of drum sounds to be assigned to particular MIDI key numbers so that your sequencer can 'play' the drums as required. A correct 'mapping' of MIDI key numbers to individual multi-samples ought to be child's play if the software's displays and use of icons (pictorial symbols) have been well thought out.
This brief article can only give a taste of the many possibilities that open up once the microcomputer can be persuaded to manipulate sample data. We know from the previous episode how MIDI can use various means - but primarily System Exclusive messages - to communicate various control and data values between the sampler and the micro. If the micro is to model the waveforms present in the sampler (so that it can display the waveforms and allow the user to manipulate the data), we must organise the transmission of the sample data to the micro.
To give you some idea of the problems this generates, let's consider a sampler that uses 12-bit data words and has enough memory to store 20 seconds worth of samples taken at 40,000 samples per second (40kHz). Each 12-bit data word is transmitted as a pair of MIDI data bytes, 6-bits to each.
In total, there will be 20 seconds x 40,000 words x 2 MIDI data bytes per word, giving 1,600,000 MIDI data bytes to transmit.
At MIDI's 31.25 KBaud transmission rate we can transmit about 3,000 bytes per second, so the time taken to transmit one full sample set will be 1,600,000 divided by 3,000, which gives approximately 500 seconds, ie. over 8 minutes. Even on the Prophet 2000/2002 sampler, which can in theory use MIDI at four times the standard speed in some cases, this would still take two minutes!
Another little problem that needs dealing with is to do with errors. It is most unlikely that 1,600,000 bytes will all be transmitted and received correctly. It is usual to split the full set of sample data into pages of (say) 256 bytes and use a method, such as a checksum calculation, to find out whether the 256 bytes of data arrived intact. If they didn't then a request is sent to the sampler, via MIDI, to retransmit just the bad 256 byte message(s) rather than the whole 1,600,000 - this is what is known in the trade as a good idea!
Once the data is available in the microcomputer, displays can be achieved without going back to the sampler for more data. When editing takes place it is sometimes possible to instruct the sampler to perform editing operations on its internal copy of the data, without having to transmit vast amounts of the edited data back to the sampler. The setting of a loop point, for example, just requires the transmission of the control value eventually selected - the actual sample data does not change either in the micro or in the sampler.
First buy your sampler and explore and experiment for quite some time. Now that you have found out what is easy, what is hard, and what you would like to be able to do more easily, start investigating the various software packages available for your particular microcomputer/MIDI interface configuration. If you are lucky, your sampler manufacturer may endorse and distribute suitable software (not necessarily their own), as is the case with the Sound Lab software for the Ensoniq Mirage sampler which runs on the Apple Macintosh. This gives you a good starting point but it is always worthwhile comparing other packages.
You should compare the graphics displays available with each package and the ease of interaction via the micro's keyboard or mouse. Don't forget to have a careful, lingering look at the owner's manual so that you can see exactly how the advertising copy translates into reality. Finally, make the salesman demonstrating the package show you how easily the software solves all of your problems: after all, if he can't find a good loop point in under half an hour what chance will you have?
Feature by Jay Chapman
Previous article in this issue:
Next article in this issue:
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!