We describe a method for the computer entry of Chinese characters using speech recognition with no vocabulary restrictions. The method exploits the fact that spoken Chinese has only about 400 syllables, and the syllable structure is extremely simple: either consonant/vowel group or vowel group alone. To each of the 23 consonants and 35 vowel groups that comprise a syllable, a code word, or Chinese phonetic alphabet representation (CPAR), is associated. All Chinese characters can be "spelled-out" using the 58 highly discriminable CPARs, thus allowing accurate transcription by means of current speech recognition technology.
The CPAR system and the speech input method were tested with a discrete word speech recognizer in an experiment with four human subjects. On average, a subject could learn the method with 22 hours of training over a period of 1 or 2 weeks. Subjects were able to translate vocally about 40 characters of Chinese text into CPARs per minute with an accuracy in excess of (97 %). This compares roughly to 31 English words per minute. The speech recognition system used was not fast enough to accommodate this rate of speaking, and in actual practice only 24 syllables, or the equivalent of 19 English words, per minute could be entered. Recently developed discrete speech recognition systems can now handle the higher rate; however, such equipment was not available at the time of the experiments reported here.