MAIN
HISTORY
MEDIA
THEORY
OTHER USES
 
SPEECH RECOGNITION

Before speech recognition, one would have to write computer code, analyze it for any errors and then compile it. The computer would then scan the written code for key words and word boundaries and from the code it would perform the commands. Using human speech as a means of conveying information to a computer provides a variety of different complications.

For example "PRINT 1 + 1" - when written - can only be interpreted in one manner, but when the command is spoken, its a different matter entirely. What if the command was spoken loudly or softly, slowly or rapidly? What if the speaker was a woman, a man, an elderly person, or a child? What about the pitch of the voice or the person's accent? Not only are there complications with the speaker, what if there are other voices in the background, or the room in which the speaker is located has very good acoustical properties? All these factors must be taken into consideration when voice recognition is used.

There is always the option of having the robot analyze the words being spoken and matching it with the closest word stored in the robots buffer. However, this will work fine for a limited vocabulary, such as "left", "right", "stop", "go". What about all the words in the English language that sound alike? There are many homonyms like "right" and "write", "night" and "knight". How is the computer supposed to distinguish the two words? Most voice functioning robots are also "speaker-dependant" and are a "discrete word recognizer", since they can only recognize the voice of one person as well as only one word at a time.

Speech Synthesis Our speech synthesis technology is much farther along than our speech recognition technology. So far there are four main techniques you can use to make your robot talk back to you. First of all, you can use a method where your robot digitizes your own voice. This method is called "pulse-code modulation." Second, you can buy a speech chip and ROM with pre-digitized voice recordings on them. Third, you can buy a speech chip that randomly generates a variety of different pre-defined words. Or fourth, you can "build" words from speech sounds, such as "phonemes" and "allophones."