AMPLIFIER NEEDS SOUND KNOWLEDGE

ZXM is a sound box from Timedata complete with amplifier. Based on the 8912 three-tone chip which appears on most sound boards it also provides a joystick socket. The Atari-type joystick socket will not work with commercial software, as it is based on an I/O port from the 8912 chip and has to be programmed to be available first. The amplifier and loudspeaker are controlled by a volume control situated on the front of the unit.

The unit plugs into the edge connector of the Spectrum and has a PCB edge at the back so that other peripherals can also be plugged-in. The edge connector is only 23 ways wide so it fits both the ZX-81 and Spectrum.

The unit is accompanied by a booklet of programs and a tape. The explanations and use of the unit tend to be over-complicated, requiring a good technical knowledge to understand it.

Timedata also makes a speech unit, ZXS. The unit, plus the software provided, allows you to set up a string (S$) of sounds to make up the words in Basic. A machine code routine will put them out to the speech unit, which also has an I/O-mapped device. Unfortunately you will be unable to hear the speech produced unless the unit is connected to an external amplifier such as the ZXM via its 3.5mm. socket.

The speech can be EDITed and changed using the program but the size of the program may make it difficult to incorporate into your programs.

The ZXM costs £29.95 and the ZXS £24.99 from Timedata. 16 Hemmels, Laindon, Basildon, Essex. Tel: 0268-418121.

SPECTRUM SPEAKERS

The word is that attaching any old speech synthesiser to your Spectrum will allow you to have cosy chats together. Henry Budgett determines where this is one of the first signs of madness.

For years science Fiction films and futuristic novels have depicted an era when man and machine can communicate in perfect harmony. The reality, of course, is slightly different. While speech recognition has yet to he fully developed (ACT's latest Rascal notwithstanding), chip-based speech synthesis has been both mastered and available for several years. Until recently the computing power needed to produce human-sounding utterances was substantial. Now almost every home computer is capable of being equipped to talk back to its owner at a price that won't even break the average piggy bank.

SOUNDING OFF

When we speak we produce three distinctly different types of sound. The most obvious are the 'voiced' or vowel-type sounds; oo, ar, ee, and so on. These are produced by air from the lungs making the vocal cords vibrate. The frequency of this vibration determines which vowel sound we hear.

The second group is the unvoiced or 'fricative' sounds; ss, sh, t and ff. Here the air from the lungs rushes past the vocal cords without making them vibrate and the frequency produced is controlled by the positioning of the lips and tongue. Finally there's silence or, to be more precise, the minute gaps that occur within words (for example six, eight) where we change from voiced to unvoiced and vice versa.

FAKING IT

In order to generate speech-like sounds, the electronics designers generally go for one of two methods. The first - and until recently the most common - is synthesis by rule. If the frequencies contained within speech are analysed it's possible to devise a system of rules that allow us to re-create any sound from its basic frequencies.

These 'building blocks' of sound are called phonemes and by using them in various combinations any word can be constructed. The individuality of a human speaker tends to be lost when speech is generated like this but the words can be clearly understood. Because the synthesis rules for each phoneme are built into the equipment, the user has simply to supply a list of phonemes to be spoken. It's then possible to generate complete sentences instantly, simply by calling up a string of stored phoneme commands. In reality these phonemes tend to be called allophones; this is because the various building blocks sound different depending on their positioning within a word or phrase. However the principle's much the same.

The second method for generating speech relies on the fact that the human ear and brain are very good at filling in gaps. The speech we hear over a telephone line is (British Telecom permitting) perfectly understandable. Yet technically the quality - the range of frequencies we can hear - is only one-fifth of what we'd expect from a standard hi-fi system. We understand what's being said only because our brain does the job of filling in the gaps.

With the fall in cost of computer memory it's now possible to convert speech into digital information compressed many hundreds of times by a wonderful mathematical technique called Linear Predictive Coding. The resulting numbers representing the original speech are stored in a ROM. To get any of the stored words out again as speech is easy; we simply give the computer the address in memory of the word and the digital information is recovered and converted back into sound, and because the original speaker's words have been stored, all the personal characteristics remain. That's why Acorn's speech chips for the BBC Micro really do sound like Kenneth Baker.

WHAT'S THE USE?

The commercial uses for speech synthesis are so many and varied that it's just about impossible to list them all. Looking just at the tip of the iceberg it can be used to replace taped announcements at railway stations and airports; in America it's widely used on the telephone system to inform callers of misdialled numbers and engaged or withdrawn services. Speech synthesis units are also being incorporated into cars like Maestros and Montegos as part of the standard instrumentation so, as well as being something of a sales ploy, they can provide warnings the driver can hear without having to take his or her eyes off the road. A major contribution to road safety perhaps?

As far as we are concerned in the home computer and electronic games market, speech synthesis is generally used to enhance games. Scores can be read out and warnings of imminent enemy attack can be given to warn players leaving them free to concentrate on the tactics of the game. Of the five speech units under review here, four of them use the phoneme system and one the stored speech method. Let's take a look at how they succeed in fulfilling their purpose.

SUMMARY

If you're looking for a means of adding a voice to your Spectrum and of incorporating the facility either into games or just for fun, then the Currah MicroSpeech is almost certainly going to be the best buy for you. It's also got the largest number of games already written for it if you prefer to use shop-bought software. Another of its clear advantages over the other units is the addition of a BEEP amplifier for putting the sound through the TV.

For those of you who haven't yet bought a joystick controller or a sound generator and fancy a speech synthesiser at the same time, then the Fuller Box/Orator combination - though expensive - offers the lot in one package.

Serious users of speech output have an equally clear-cut choice. The superior quality offered by the DCP S-Pack's Digitalker chips make this the logical buy for anyone using the Spectrum as an annunciator rather than as a games machine. The manuals supplied aren't good enough by far, but the Digitalker chips are more versatile than you might think, so if you buy this one get in touch with National Semiconductor for the real data.

Of the remaining two units, the Cheetah offers a built-in amplifier and speaker whereas the Timedata unit doesn't; their respective prices reflect this. Neither of them comes close to the overall 'usableness' of the MicroSpeech and they both lack the BEEP amplifier and keyword voicing.

ZXS SPEECH SYNTHESISER
Price: £24.99
Timedata
16 Hemmells
Laindon
Essex SS15 6ED

HARDWARE: Physically this is the smallest of the review units at just 65mm by 70mm by 40mm. Housed in a vertically mounted 'potting' box it offers an expansion bus connector and uses the GI allophone chip.

The unit is too small to include a loudspeaker so a 3.5mm jack socket is provided for connection to an external amplifier and speaker. Construction on the single PCB is neat and well thought out.

SOFTWARE: Two programs are provided on the accompanying cassette. but the second one - a speech editor from the manual - stubbornly refused to perform, giving 'out of memory' errors when trying to RUN. The cause was an over-dimensioned array, but it's a bug all the same.

You can access the allophones direct by using their codes and doing an OUT, or you can make use of the machine code subroutine provided and build strings of allophoncs in the reserved variable s$. In this respect the unit works in much the same way as Currah's MicroSpecch, but without the advantage of having the software built-in.

MANUAL: Twenty A5 pages cover the theory and practice of allophone synthesis and example programs are included as well as being supplied on tape.

SUMMARY: Although it's the cheapest of the bunch, the lack of an internal amplifier and speaker is a nuisance. With these added it would have been directly comparable to the Cheetah.