摘要 |
PURPOSE:To prevent a decrease in recognition rate due to an error of an input phoneme or syllable and a decrease in real-time response by providing analyzing the cause of the error and providing a word detecting means and a sentence detecting means which utilize its information. CONSTITUTION:An error cause analyzing means 500 which analyzes the cause of the input phoneme or syllable as an identification error caused when the phoneme or syllable is identified as the feature parameter of an input voice and its analytic result is used as information on word detection 300 and sentence detection 400 to detect the detecting operation. In the word detection 300 and sentence detection 400, the determination accuracy of error patterns of a phoneme string or syllable string as a phoneme string or syllable string as a word candidate is therefore improved. Consequently, the recognition characteristics of a continuous voice are improved and recognition processing is performed nearly in real time. |