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.
Supplier: William Stuart Systems or Commotion
Price: £49.95 and also requires Chatterbox module or WSS I/O port.
Useful and fairly effective speech recognition unit. Will mainly be of use to the disabled, educational establishments or collectors of computer gadgetry.