| ★ HARDWARE ★ AUDIO - SYNTHETISEUR VOCAL SSA-1 ★ |
| Synthetiseur vocal Amstrad SSA-1 (Microstrad) | Synthetiseur vocal Amstrad SSA-1 (Amstrad Magazine) | Amstrad Speech Synthesiser SSA-1 (Amstrad Computer User) | SSA-1: Amstrad Speech Synthesizer (CPC Magazin) | Amstrad Speech Synthesizer (Home Computing Weekly) |
Let your Amstrad do the talking... Not so long ago, synthesized speech of any kind wasn't really feasible, even on the largest of computers. The extraordinary speed of research and development in the computing and electronics field now makes speech commonplace, and almost trivial, even on the humble home micro. Everything you need can now be obtained on a single chip, the SP0256, costing just a few pounds. All you have to do is interface the thing to the computer of your choice, amplify its output and hook up a speaker. Now two companies have done just that for the CPC464.
DK'TRONICS SPEECH SYNTH First on the speech scene were peripheral experts DK'tronics , whose speech synthesiser was announced a few months ago as being the first add-on for the Amstrad. Production seems to have lagged, though, and it is only now that their synth is becoming available. The main unit is a small charcoal box about four inches wide by three inches high and an inch-and-a-half thick. It plugs into the main expansion port on the back panel and has a through connector to allow the connection of other peripherals. The styling is excellent: the cross-section through the case is deeply angled so that it fits tightly against the contoured back of the 464, and the ventilating slots along the back of the computer are duplicated too. It looks more at home than Amstrad's own peripherals, and doesn't flex about in the same way as the floppy disc interface, despite the latter's contoured ribs. The speech unit will also amplify the Amstrad's own sound output, in true stereo. It obtains the signal via a jack which plugs into the socket next to the joystick port. The twin speakers, which appear to be car stereo types, plug in either side, and an overall volume control is provided. Stereo balance is factory-set, just as well since the balance test program in the manual doesn't work. Connecting up the unit revealed two problems. First, unless you perform surgery, you can't use the DK unit with the excellent Arnor ROM board reviewed last month. This is because DK have fitted the polarising key to the connector, but Amor haven't cut the corresponding slot in their board. And Amor's board can't go in the back of the DK unit because it's too deep. One solution is to pull out the polarising key with a pair of pliers, though this may cause the speech unit to slide along the edge connector and make poor connections. A better, though trickier, method is to get a fine hacksaw or fretsaw blade and cut a slot in Amor's board in exactly the same place as the one on the Amstrad edge connector. The second problem is that the DK unit draws a lot of extra current. I couldn't use the unit and my floppy disc at the same time because the overload protection in the monitor shut down the power supply. I'm told that this only affects some Amstrads, but if you're one of the unlucky ones I don't know what the solution is. The method of speech synthesis used by the SP0256 is allophone generation. Allophones are the phonetic building blocks of speech, the basic sounds that we string together to form words. Including a range of five pauses of various lengths, there are a total of 64 allophones. These are numbered, 0 to 63, so interfacing is simply a matter of providing a single I/O address for the chip. Speech is generated by writing a stream of single bytes (the allophone code numbers) to the chip, with handshaking via the same address. So, we need a little software to control the flow of allophone numbers to the chip, and a lot of software to control what's actually being sent. DKTronics supply this on a cassette as a set of RSXs (resident system extensions, or I commands), so the facilities can be used equally well from BASIC or machine code. The software asks you where you wish it to be located: suitable addresses range from 16384 to 39000. There is no default: you must type in a number. HIMEM is lowered and the RSXs installed above it. There are a total of six. The software uses two buffers: one to hold the words, the other to store the allophone data to be output under interrupts to the chip. |SPON and |SPOFF are used to turn the interrupts on and off: this merely halts the transmission of data, so the buffer contents are left intact and the last allophone will sound continuously. |FLUS kills the speech and flushes the buffers completely. If you are using the text-to-speech software, there are three modes, set by |OUTM plus a parameter. Mode 1 will only sound text which is in a PRINT statement and enclosed by 1 marks (inside the normal " marks). Mode 2 says all printouts including listings, error reports and 'Ready', but without the text appearing on the screen. Mode 3 says everything and lets it appear on screen too. The speed of the speech can be controlled by |SPED (why all these abbreviations to four letters? An RSX can have a command of any length). This doesn't affect the actual pitch of the voice, merely the time taken to sound each allophone. The range is 0 to 15, but only the smaller values are useful: at the slowest speeds, above 10, the allophones don't flow together very well and the speech is harder to understand. Note that |SPED acts asynchronously with the data in the buffer: if you send some text to be said at one speed, then swap to a new speed and send some other text, the speed command acts instantly and all the data in the buffer comes out at the second speed. The text-to-speech tables are good, but not infallible. In some cases you will want to send allophones directly to the chip to get exactly the right pronunciation, and the command for this is |FEED. To give you an idea of speech construction, the full table of allophones is provided plus a two-page dictionary. This has some really useful words, like bathe, cookie, sweats and threaders. Some of them aren't very good, either. DK suggest 16,7,7, 52,19,43 for 'memories'. This comes out like 'me-riss'. I played around and decided 16,7,7, 16,23,58,12,12,12,43 sounded much better. Experiment a little: you'll find phonetic construction easy with a bit of practice. THE AMSTRAD SSA1 Amstrad's version comes in a box very similar to the floppy disc interface moulding, except that it's a couple of inches shorter. It plugged into my Amor board without difficulty and drew a small enough current to allow operation with the disc drive connected. Apart from that, things are much the same as on the DKTronics unit. The internal stereo sound is fed into the unit via a jack plug from the sound port, and the twin speakers, again of car HI-FI design, plug in at either side. A master volume control is fitted on the right hand side. Again, only a single I/O address is required, but Finagle's Law dictates that Amstrad have chosen &FBEE while DKTronics use &FBFE. This might upset the software houses: if they write games with a speech option, the software has to muck about testing which port is in use and alter itself to suit. How about a bit more co-operation in future, guys? As always, the documentation that accompanies the hardware is a superb production number. Twenty A4 pages tell you everything you ever wanted to know about speech synthesis. Once again, control of the speech unit is made easier by a set of RSXs (completely relocatable) supplied on tape. It's perhaps surprising that neither company included the software as external commands in a ROM, since that was the design philosophy behind peripheral handling on the Amstrad. DKTronics quoted lack of R & D time as the reason for their choice, while Amstrad said they did it to keep the price down. Fair enough, but once you've tasted instant software on ROM, anything else is a bit of a let-down. Nevertheless, the RSXs total nine, and provide a pretty comprehensive selection. The software occupies just over a kilobyte, compared to the 4K needed by DKTronics, as the SSA1 uses a text-to-speech algorithm (set of rules) rather than tables. This means it gets fooled on some everyday words, even if you try to help it out by altering the spelling. For example, to take leave of our editor a suitable phrase might be 'Goodbye Simon', but this comes out as 'gooodbee simmon'. 'Guudbiii Seyemon' is better, but as with the previous unit, sometimes the only answer is direct allophone generation. On the SSA1, this is done by |APHONE. Text-to-speech can be performed using embedded 1 marks in PRINT statements, though the syntax here is |ECHO plus a mode number for all text, listings only etc. |SAY allows text-to-speech without the hassle of including the reverse quotes. The way in which text is spoken is rather better on the SSA1, in that numbers are spoken as we would pronounce them and not as single digits. For example, if a listing is being spoken aloud, the SSA1 says 'ten' and 'twenty' for line numbers, whereas the DK unit says 'one zero' and 'two zero'. Again, |SPON and |SPOFF enable and disable the interrupt-driven processing of the allophone data in the speech buffer. |QUIET is the command to flush the buffers and silence the chip. However, the Amstrad software also provides three extra commands for low level driving of the hardware. The first is |ROOM, which interrogates the system and tells you how many free locations there are in the allophone buffer, whether an allophone is being sounded, and whether the speech interrupts are active. |SPSTATUS is similar but only returns information about the hardware status of the chip itself. Finally |SPOUT is not a cue for the chip to talk incessantly, but sends allophones directly to the chip without passing through the buffer. You have to disable speech interrupts and control the handshaking yourself here. There are a lot of useful demonstration programs included, such as a BASIC speaking clock which announces the time every five seconds, and tips on how programs should test for the presence of either the hardware or the SSA1 software. CONCLUSIONS As far as features go, there isn't much to choose from between the units. The DK device has a better styling and appearance, and has a much bigger phonetic library. On the other hand, the SSA1 has more commands and can provide a reasonably accurate text-to-speech conversion using less memory space, which could be critical on some of the largest commercial games. The actual sound of the speech has a different quality on the two systems, but it's difficult to put into words exactly what this difference is, or which you might prefer. I think it only fair to point out that the chip has to do a lot of approximating to enable such a complex thing as speech to be duplicated at all, so neither synthesizer sounds any better than your average Dalek. I think it will be some time before we find blind people programming on the Amstrad and having their listings read back to them coherently: the main problem is that syllables are all given equal emphasis. It's surprising how difficult this can make comprehension. However, the units will allow an extra dimension to be given to games programs. Peter Green , ACU |
|  |
