摘要 |
PROBLEM TO BE SOLVED: To obtain a front-end processor which can perform speech recognition processing fast and is inexpensive and easy to use by recognizing a speech signal in phoneme units, and recognizing a phoneme or a string of phonemes while altering it as character information. SOLUTION: A voice inputted through a microphone 1 is converted by an A/D conversion part 2 into a digital signal, which is supplied to the front-end processor 3. The front-end processor 3 processes the inputted digital signal to take an acoustic analysis, thereby extracting feature parameters. Then, the extracted feature parameters are applied to a phoneme model of HMM and the phoneme is recognized. Then, the time series of phonemes is used to find candidates for plural words and the word having the highest transition probability among phonemes is selected and recognized out of them. Further, the recognized word is verified by using a grammatical model to judge whether or not the word is accurate from language information on document units, and the word is displayed as a recognition result at a display part 6. |