Big Ears

Sinclair User

Overall 2/5

John Gilbert 2/5

Supplier: William Stuart Systems or Commotion Price: £49.95 and also requires Chatterbox module or WSS I/O port. Big ears is a CB term for operators who listen to conversations which don't concern them. Nothing could better describe the speech recognition package from William Stuart Systems. The package contains the SRI speech recognition unit with connector DIN cable, a battery powered microphone and a demonstration program. Speak into the microphone, teach Big Ears various words, and it can learn to recognise them. Big Ears connects to the Spectrum by either the WSS Chatterbox speech synthesiser or the WSS I/O (input/output) port - a review of the latter is included elsewhere in this feature. Big Ears only stores and recognises one word at a time and holds a maximum of six on a 48K Spectrum. First, give the word a number by which it can be identified in the Big Ears Ram dictionary and spell your chosen word. Once you've done that the program prompts you to press Enter and speak the word into the microphone. When you press Enter you'll have approximately five seconds to speak the word. Big Ears takes four samples of your voice of each word and translates each sample into a voice print which is an array of 36 numbers. The computer takes several samples of your voice saying the same word because, just as everybody has a different voice, so your voice is slightly different in level and inflection every time you say the same word. The four computer voice prints are averaged to give a range of levels and inflections at which you might say a specific word. The result is not always accurate. You may, for instance, want the computer to recognise the word 'science' but it keeps comparing it with 'psychology' which Big Ears has also learned. The reason is that you - and most other people - pronounce 'psy' and 'sci' in a similar way. Big Ears has no real auditory sense and. If there are several like - sounding syllables in certain words it has learnt it will get confused and maybe respond with the wrong word. When you've taught the computer two or more words you can Test and Demonstrate speech recognition Type T, for Test, if you want to see the calculations which the system does to match your word with one in its dictionary. Speak into the microphone when prompted by the program and the correct dictionary entry should be displayed on the screen with a voice print and correlation table. The Demonstration option is similar to Test. Big Ears can make your computer seem intelligent. Chatterbox can respond vocally to your spoken commands into Big Ears. Alternatively, you could use the package to control robots with spoken commands through the I/O port. For instance, you could teach Big Ears 'left' and 'right' and write a Basic program to link those words to the movement of a floor turtle or similar robot. Instead of typing in the word which corresponds to your spoken word you could type in a reply to a command. For instance your spoken word could be 'up' to make a mechanical arm move up. The reply could be 'your command is being obeyed'. Type in the reply and teach the computer the word. The Big Ears control program was written in Basic so that you can alter it, add extra lines, change the number of words the dictionary holds and pull out options that you may not require in your programs. The speech routines, however, are fully machine-coded. Instructions on how to customise the program and add routines are given in the short booklet which accompanies the package. It also includes line-by-line details of the control program so that you can adapt it to bring other devices, such as robots, security systems, or even the infamous Red Box at your beckoned call. Both Big Ears interfaces are open-ended so you can add whichever devices you require on to the back of your Spectrum and control them through the machine's ports. For the majority of Spectrum owners, Big Ears is an expensive show-off rather than a practical show piece. It has an irritating habit of picking up background noise making it useless in any environment other than a padded cell. If you are, however, desperately seeking new things for your Spectrum, Big Ears will keep your interest for a couple of hours at least.

Personal Computer News

John Lettice

ITEM: Big Ears PRICES: £49 MACHINE: Various CONTACT: William Stuart Systems, [redacted] MICROS HAVE EARS You can talk to your Spectrum, but you won't get a reply - John Lettice looks at Big Ears. Speech is still something of a new frontier as far as the cheaper micros are concerned. The problems that must be solved are complex, but some of the tools you need are now available. PCN looks at a duo from William Stuart Systems. Speech synthesisers are now so common that the world and his wife's micro seem to have access to them. In the case of mass-market micros like the Spectrum, the range is so wide that users will encounter considerable difficulty making a choice. Assuming one Dalek sounds pretty much like another, you may think that making a choice is simply a matter of the convenience and prices of the various pieces of hardware. But when you think about the sort of things you'll want to use a speech add-on set-up for, you'll realise that there are other considerations. Speech synthesis at the low-end micro level has been conquered to all intents and purposes, but the problem of cheap speech recognition still remains. Wouldn't you like to have full verbal control over your micro, with all that implies for control of peripherals? For example, you could stand in the middle of the room bossing a turtle around - and at the really gosh-wow end of the market, you could phone up your ZX81 and tell it to switch on the video recorder. Interested? Well, the good news is that it can be done - the bad news is that it can't. The reason things are a bit equivocal at the moment is that, while the hardware that makes possible the solution to the problem exists, the software to make it work still has some way to go. For example, William Stuart Systems is currently marketing a speech synthesis unit called Chatterbox, and a speech recognition unit called Big Ears for £49 each, giving you the necessary hardware to - in theory at least - produce a very cheap speech system. Compare the price with the £899 for the Apple Voice Input Module (PCN, issue 15) and you'll naturally assume there's a snag. Well there is - Big Ears will recognise your speech quite effectively, but you'll have grave difficulties getting it to produce more than demonstration speech output, because you'll either have to write your own software, or wait. PRESENTATION Big Ears can be connected to a wide range of micros through the Chatterbox speech synthesiser or through a number of commercially available input/output ports. The test model was run on a ZX Spectrum through the Chatterbox, and came with a microphone, lead and driver/demo software tapes. We tested it with Chatterbox 1, which can be upgraded to Chatterbox 2 for £5.75. The whole ensemble came in a tidily packed box, complete with copious quantities of instruction leaflets. DOCUMENTATION The presentation of the instruction leaflets is, to put it mildly, a shade on the rudimentary side. Chatterbox is well documented, but the main manual for Big Bars (15 pages of stapled A5) seems to spend most of its time on the UK101 and Nascom. Along with the main manual you get two sheets of A4 giving supplementary information for the Spectrum and ZX81. If you're running Big Ears off a ZX81, the POKES you need are given, but the rest of the machine-specific information seems to be just a variant on the Nascom/101 instructions IN USE Getting the whole set-up running was relatively easy. You plug Chatterbox into the Sinclair's edge connector, connect it up to Big Ears, plug the microphone into Big Ears, power up and you're away. Load up the Big Ears software, and you can begin teaching your micro to recognise words. You're initially asked whether you want to 'learn, test, demo or save'. Your first move, logically, is to select '1', which allows you to place a word in the machine's vocabulary. Each new word should be assigned a word number, and then the word itself is typed in. You then press Enter and say the word, four times, by which time the Spectrum has a fair idea what you're on about Once you've got a couple of words stored - with the software supplied the maximum is ten - you can test Big Ears' recognition capabilities. At first I had problems here - I'd hit t, then Enter, and as I was gathering breath to say something, 'You said pig' would flash up on the screen. The manual was silent on this subject, and the main manual itself gives the impression that you'll get a 'Please speak ... Now!' prompt before having to open your mouth. However, I tried bellowing at it the moment I'd entered 't', and found I got the required attention. The recognition seems quite good for one voice, but when we tried varying the voices it learned from, it showed a tendency to confuse Igor and pop If as is very likely you find you need to store more words, instructions are given on how to alter the program to increase its capacity. You can also vary the number of repetitions needed for the program to store a word, bearing in mind that the more repetitions, the better the recognition. Round about this point you'll start wondering what you could get it to do next, and how. Big Ears stores words as a sort of digitised voiceprint, and you can see this as you teach it words, should you wish. But Chatterbox, and most other cheap speech synthesisers, work on an allophone system, allophones being units of sound that human speech is made up of. The problem is turning your voiceprints into something that the Chatterbox can reproduce in allophone form if, that is, you want to be able to talk to Big Bars and have Chatterbox repeat your words. But note that this is not the same as Big Ears understanding you. What you need to make the device of more immediate use is the software to allow it not simply to repeat words but to act on them. Now, provided you can produce a stable enough sound, and back it up with software that will recognise the 'print' of that sound - obviously not an exact print, but one within a given margin of error - you could get your micro to execute an action. As the demonstration software included with Big Ears allows you to alter the number of repetitions the device demands before it 'learns' a sound, and as the more you do the more accurate it is, you can probably nail its print down fairly tightly, and bear in mind you don't have to stick to sounds you make with your larynx. No, don't go away... if what you want is a series of simple actions, why not try clapping, a knock on the desk, whistling? Why not be the first on your block to have a micro that dims the lights every time you clear your throat?. VERDICT The potential for Big Ears is clear, but unfortunately it's still largely unexplored. William Stuart Systems is currently working on applications software for the Spectrum, and UK101 software is probably a little further advanced, but the company says that, because users will tend to want to employ Big Ears for specific applications, it's very difficult to decide what sort of routines to include. So, really, what you're talking about here is a product that needs a fair bit of programming input before you can use it to its full potential, and if you want to control your environment by voice, you'll also need to be pretty nifty with a soldering iron. But this isn't a complete disqualification for novices. If you're willing to teach yourself, and if you're clear in your mind that it will take you some time to get the results you want out of Big Ears, then it does represent excellent value for money.

The complete set-up - Big Ears, top left, is linked up to the Spectrum through Chatterbox. Jut speak into the mike and your voiceprint appears on the screen.

The inside of Big Ears consists mainly of a commodious voicebox.

TWO-WAY COMPUTER COMMUNICATION The Chatterbox - also known, as you can see, as Speech Synthesis by William Stuart Systems - will connect direct to your micro to provide an allophone base speech synthesis. Speech recognition - nom de guerre Big Ears - links up through the Chatterbox, and renders your speech into easily recognisable voiceprints. Allophones may take a little longer.

ZX Computing

ADDING ON YOUR SPECTRUM Our review team take a brief look at some of the hardware add-ons fir the ZX Spectrum. Now that the ZX Spectrum has well and truly established itself on the micro market, it is amply supported by a wealth of hardware add-ons. Most of these peripheral devices have been manufactured by the people who supported the ZX81, but just as the Spectrum has attracted new users with its prowess, so too has it attracted a new following from the add-on manufacturers. In this brief guide, we have not tried to cover all the devices currently available on the market but rather give you a flavour of the technology you can add on to your Spectrum. As joysticks are an obvious favourite, these have been covered in some detail whereas RAM packs are fairly standard and so have only been briefly touched on. Also included in this section area number of quite specialised add-ons like sound units and a digital tracer. If you own a Spectrum, you will no doubt have begun to realise the potential you hold in your hands. Over the next few pages you will hopefully see further applications for you and your computer to explore. BIG EARS - WILLIAM STUART SYSTEMS Originally designed for the UK101 and Superboard, the Big Ears speech recognition system is now available for most leading micros including the ZX Spectrum. Housed in a sturdy box, the Big Ears system consists of a microphone, pre-amplifier, analogue frequency filters and digital interface. You are also supplied with a software package allowing you to become acquainted with what you soon find is a very complex piece of technology. The program with the system has four modes. Learn, Test, Demo or Save. All the modes are fairly self-explanatory, and the demo explains all should you require comprehensive details. Using the equipment, you can allow the Spectrum to learn a word, you can then test that the computer understands and recognizes that word, and then save the program together with its new expanded vocabulary. During the review period, we had the Big Ears programmed to recognise nearly everyone's name in the office. And this it did quite successfully, except for the name 'Helen' which it seemed to 'recognize' every time it wasn't quite sure! Complete with full instructions for use, the Big Ears speech recognition system is priced at £49. For further details of this device, contact William Stuart Systems Ltd, [redacted].

The Big Ears speech recognition interface from William Stuart Systems.

ZX Computing

NOISY LITTLE THINGS In this issue, our hardware reviewers look at a system which allows your ZX81 to listen, a ZX81 sound maker or three and a device to ensure that you don't lose a program when the power fails. ZON SOUND UNIT A wide range of sound effects can be added to your ZX-81 with the ZON X-81 Sound Unit now available from BI-PAK. The unit is based on a three-channel-plus-noise sound chip, and is so designed that the pitches and volumes of the three channels and the overall attach/decay envelope can be controlled by simple BASIC statements. By this means, piano, organ, bells, helicopters, lasers and explosions can be simulated and easily added to existing programs. ZON X-81 is completely self contained in a neat black plastic case with loudspeaker and manual volume control (in addition to programmed volume), and simply plugs in between the rear of the ZX-81 and its RAM pack and/or printer (if fitted). No dismantling, wiring, soldering, batteries, power supplies or leads are required. Instructions take you through the operation of the unit step by step and include a number of example program of useful sound. It is available from BI-PAK Semiconductors, [redacted]. Price: £25.95, including postage and V.A.T. BIG EARS The 'Big Ear's speech recognition system, which costs £49, plus VAT, including p&p, consists of a microphone, pre-amplifier, analogue frequency filters and digital interface, complete with software. Words are stored as voice patterns which the system learns from repetition by the user. Ten or so words can be stored at a time. The computer then checks each word spoken against its word bank, and assigns a percentage to each word regarding its chance of being the word spoken. It assumes the word gaining the highest score is the spoken word, and responds accordingly. It is available from William Stuart Systems, [redacted]. PROTECTING THAT PROGRAM The 'software protection unit', made by Microbyte of Lichfield, looks like being a partial answer to all those horrifying moments when 8K worth of carefully typed in programs suddenly vanishes. The software protection unit has a socket to take the mains power unit jack plug from your ZX81 mains transformer, and the plug on the unit then hooks into the ZX81. If you unscrew the four screws on the lid of the unit, you'll see two battery containers, which hold a total of six batteries. These are the secret of the unit, which is designed to take over if the power fails. When we tried it, we found it worked exactly as described. We didn't try the 'how long will a program last' test, but take the manufacturer's word that with a ZX81 alone, a program will stay intact for three hours, and with the 16K attached, about an hour and a half. You're cautioned to always remove the power jack plug on the ZX81 after turning off the mains supply, or the batteries will think there has been a power supply, and wear themselves out keeping the '81 going. The batteries are, of course, only intended for emergency use. Excessively used batteries are signalled by video character deformation, and loss of sync, although at this stage the RAM contents are not lost. The unit costs £8.60 and is available from Microbyte, [redacted].

The Microbyte Software Protection Unit

William Stuart-Bruges talks to his attenitve ZX81 through the 'Big Ears' speech recognition system.

Sinclair User

ZX81 GETS BIG EARS The Big Ears speech recognition system is available for the ZX-81. The system consists of a metal box containing two frequency filters, a battery-driven condenser microphone complete with stand, and a DIN socketed cable to connect to your iNPUT port. The only ports recommended by the firm are Technomatics, its own synthesiser board and the Redditch port. Other ports could be used, but their address would have to be inserted in the machine code routine used to obtain the voice from the frequency filter. The system expects to have 16K of RAM attached to it and the port must be out of that section. The program reserves 256 bytes at RAMTOP for its own use and it is mainly a Basic program. The connection to the ZX-81 port must be done by soldering it to the port, as the DIN plug fits on to the frequency filter. The system can be used to store voice prints, which are used to control the actions of the program. So shouting LEFT-a-bit, DOWN-a-bit could become common. Big Ears, costing £56.35, can be obtained from William Stuart Systems, [redacted].

Big Ears