Published in Acorn User – April 1986. A test of (then available) 3 systems for speech synthesis.
Please note: all (c) by the Author Martin Phillips and Acorn User magazine.
One of the most innovative features of the BBC micro was, and still is, the provision for speech synthesis, yet it has been virtually ignored until recently. The Acorn speech synthesiser has not achieved great acclaim, and several other inexpensive systems that have been produced have proved almost unintelligible except to the practised ear. Recently, two interesting developments have come along – Computer Concepts’ Speech ROM which’ uses the on-board speech synthesiser but extends its vocabulary without a great loss in quality, and Cambridge Microcomputer Centre’s Namal Type and Talk speech synthesiser which requires almost no programming. Speech can be recreated by recording certain words and processing them so that they can be This method can achieve a high quality of reproduction but you are limited to the words already stored in the system. It’s not a perfect system, because words have various inflections when used in different parts of a sentence. To do this system justice, phrases or sentences would have to be recorded, thereby increase problems of storage and access. Th realistic where the application is few sentences, such as in cars whcr systems are used to warn of faults. Human speech is considered to practical upper frequency of 4kHz. store speech, digital sampling of the signal is required to be performed at a rate of 8kHz. Each sample will need to be resolved to 12 bits, which gives an approximate data rate of 100,000 bits per second! In more practical terms it takes 20 seconds to utter six English phrases averaging six words each. This implies a memory storage of two million bits of data. A BBC micro would be able to store just two seconds of speech using all its available memory at this rate. To make speech synthesis a practical proposition, therefore, special techniques have been devised to reduce the amount of memory needed to store a sound – this is the way the BBC speech system works. The alternative is to break English speech down into its component parts (phonemes) and then reassemble them to make up words. There are just over 40 phonemes in ordinary English, and of course other languages such as French or German will have different sets and numbers. The advantage of the phonemes system is that any word can be constructed by putting together these voice sounds. The disadvantage is that the words don’t sound real as it’s difficult to give the phoneme the inflection its position in the word demands, and so the result sounds rather artificial. One is left with a compromise, either to use a limited vocabulary of high quality speech, or to have a wide vocabulary of rather stilted speech. Here one must examine the reasons for wanting speech on the computer before making a decision on which system you want. In the near future, speech synthesis will be used to give audible instructions or warnings on machines or equipment. These applications demand a system that has a limited vocabulary, yet is easy to comprehend. On the other hand, a wide vocabulary is required, say, to use the computer as a spelling tester, but the speech would also have to be very clear. Provision for speech has been made on the circuit board of the Beeb. There are two vacant 28-pin sockets near the keyboard close to the power supply, designed to hold the Texas TMS 5220 voice synthesis processor and the Texas TMS 6100 voice synthesis memory. The upgrade for the spee_sh synthesis system comprises these two integrated circuits, another multi-way lead from the main circuit board to the keyboard circuit board, a socket on the latter board and a cover positioned at the left-hand side of the keyboard for access to the socket. One purpose of the socket is to allow extra speech memory modules to be plugged into the computer. Once installed, the speech system uses an existing Basic statement, SOUND, and does not affect the normal working of the computer. Early . computers (issue 3 boards) require some modification to them before the system will function correctly – the instructions are provided with the kit, but it’s not an easy job. Issue 4 boards on do not need any modification and it’s an easy job once you find which chip goes where, as the documentation is not clear. The synthesiser is simple to use, requiring only straightforward commands to produce spoken words. The speech system has been modelled on the voice of the BBC broadcaster Kenneth Kendall, and the quality of speech is very high. The disappointment of this superb, but very limited system, is that Acorn has never introduced more speech modules to increase the vocabulary, and make the system more useful. The system is designed so that different memory modules can be accessed. The Acorn system is limited to a vocabulary of 165 words and part words, which allows few applications – really only useful for number work, as it can be programmed to count correctly up to one million. The system will say each letter of the alphabet, so you might think it’s a superb system for teaching children the alphabet except that it says ‘ay’ instead of’ah’ for ‘A’ etc. At the moment there is very little software to support the Acorn speech system. The Computer Concepts Speech ROM is really exciting. It require_s the on-board Acorn speech processor, the TMS 5220, but not the associated speech memory ROM. The Computer Concepts ROM drives the speech processor directly, and is even able to adjust itself to cope with the fault on early circuit boards already mentioned. The Speech ROM needs two pages of memory to operate, and therefore it has to be enabled before use. This is a much more sensible system than, say, the Econet chip which grabs its memory and then you have to claim it back if you are not using it. The Speech ROM comes with a 64-page handbook which is quite easy to follow, although a few more simple examples at the start would be helpful. It’s quite straightforward to program and there are lots of examples to illustrate some of its very advanced features. The quality of sound 1s not quite as good as that of the Acorn Speech system, but is still very good, and there is no difficulty tn understanding what has been said. I was most impressed with the error message system which makes debugging much easier. All in all a cheap, simple system that is easy to program, with a high quality speech output. The Namal Type and Talk speech computer is very different from the other two systems. For a start it’s a self-contained speech unit with its own amplifier and speaker, but its major feature is that it does not need programming. It will ‘say’ any words sent to it, in much the same way any words sent to the printer will be printed. This, therefore, has the advantage that once the speech computer has been enabled, it does not need any programming commands to speak any words sent to the screen. The speech computer comes with both serial and parallel input, and either can be used with the BBC micro. First you must plough through all the details of DIP switch settings in the handbook. I wonder how many Beeb users know whether_ the serial port has odd or even parity, let alone how many st~p bits? Having worked this out and checked the switch settings (they were actually set for a BBC micro, but there was no note to say so), I tried to discover how to use the system. It’s actually l’ery easy but you would never know that from the manual! In the end I struggled for some time, checking and rechecking, before finding the connecting lead was at fault, and not the DIP switches or the programming. With the serial printer port enabled, and the printer option on, the Type and Talk repeated everything that was printed on the screen. Its ‘voice’ is electronic, but usually intelligible. Such a simple system is forced to have its drawbacks, and this one is no exception. It translates each sequence of letters into recognisable words, but the English language is not to be tamed so easily. The Type and Talk cannot differentiate between words spelt the same but pronounced differently (eg, wind), but it does a creditable job in most circumstances. It’s possible to alter the pronunciation of such words by programming it first. Of the three systems, the BBC speech system at its full price of £55 is expensive, although it does offer high quality speech, but its limited vocabulary restricts its application. The Computer Concepts SpeechR OM priced at £33 (plus £10 for the BBC speech processor ROM, if required) is a far more versatile system than the BBC alone. Its quality of speech makes it suitable for many uses, for example in education ( especially for infants who are unable to read screen instructions), with handicapped people, in programs with overcrowded screens, or in software where it is inconvenient to watch the screen all the time. The Namal Type and Talk speech synthesiser is expensive at £171.35, and its handbook needs to be rewritten. The quality of f ‘ TheN amaTl ypea ndT alkn eedsn op rogramming speech is not as good as Computer Concepts’ ROM, but it can be used with little programming effort and knowledge. This lends it to a variety of applications – again, use with handicapped people comes to mind, where it would be expensive to rewrite existing software to use, say, the Speech ROM. The lack of quality, however, limits the range of possible uses – it’s not good enough for infants. Speech has been one of the most underdeveloped areas on the BBC micro – largely due to the very limited facilities of the BBC speech system and the poor quality of many of the alternative systems – and yet it holds considerable potential. The Namal Type and Talk adds a new dimension as it is an easy to program system that can be implemented with almost any software. The Computer Concepts’ Speech ROM gives an on-board system at a reasonable cost and its quality is more than good enough for a wide variety of applications. This is the system I think will have the most potential. Now the tools are available, we shall have to wait for applications and software to be developed to see how successful these systems will be in making effective use of speech. BBC Speech System, £55 Available from Acorn dealers. Computer Concepts Speech ROM, £33 + £10 for the speechprocessor if required Computer Concepts, Gaddesden Place, Hemel Hempstead, Herts HP2 6EX. Namal Type and Talk speechcomputer, 171.35 Cambridge Microcomputer Centre, 153-154 East Road, Cambridge B11DB. 183