| 摘要 |
<P>PROBLEM TO BE SOLVED: To determine conversion candidates for a character string wrongly recognized by speech recognition processing on a basis different from that of the speech recognition processing. <P>SOLUTION: A speech recognizer comprises: a speech data acquisition unit for acquiring speech data; a speech recognition unit for performing speech recognition processing to the speech data in reference to an acoustic model and a first language model, and creating a recognition character string indicating the recognition result; a conversion target character string determination unit for determining a character string included in the recognition character string as a conversion target character string; a reference character string determination unit for determining a character string connected to the front or back of the determined conversion target character string in the recognition character string as a reference character string; a conversion candidate character string determination unit for determining, in reference to a second language model, a character string whose connection relationship with the determined reference character string is indicated as a conversion candidate character string of a candidate converting the conversion target character string; and an output unit for outputting the determined conversion candidate character string. <P>COPYRIGHT: (C)2012,JPO&INPIT |
