| 摘要 |
PURPOSE:To decrease the number of 'KANA' (Japanese syllabary) candidates and to improve the correct conversion rate of a 'KANA' string by referring to past recognition results. CONSTITUTION:A speech recognition part 2 recognizes an input speech from a microphone 1 and sends it to a linguistic processing part 3. The linguistic processing part 3 generates plural syllable lattices based on recognition results outputted by the speech recognition part 2. A 'KANA' string generation part 32 judges a recognition result to be used by using both past syllable lattices and current syllable lattices to generate a 'KANA' string candidate, which is supplied to a 'KANA'-'KANJI' (Chinese character) conversion part 33. The 'KANA'-'KANJI' conversion part 33 inputs a 'KANA' string candidate supplied from the 'KANA' generation part 32 and requests the 'KANA' string generation part 32 to send a next 'KANA' string candidate while performing 'KANA'-'KANJI' conversion processing by using a 'KANA'-'KANJI' conversion table in a dictionary part 30. A 'KANJI'-'KANA' string obtained by the 'KANA'-'KANJI' conversion is supplied to a document display part 4 and displayed. Conversion is canceled through a control part 5.
|
