Speech recognition system and speech recognition method

A speech recognition system comprising:

• a context storage medium storing a monosyllabic command context including a plurality of monosyllabic commands and storing a polysyllabic command context including a plurality of polysyllabic commands;
• a speech segment detector detecting a speech segment having a start point and an end point by analyzing a speech signal within the speech segment;
• a syllable determiner determining whether the speech signal corresponds to a monosyllabic form or a polysyllabic form;
• a feature vector extractor extracting a feature vector by analyzing the speech signal;
• a speech recognizer calculating initial confidence scores of respective monosyllabic commands and initial confidence scores of respective polysyllabic commands based on the extracted feature vector;
• a weight value setter setting a first weight value which is to be applied to the initial confidence scores of the respective monosyllabic commands and a second weight value which is to be applied to the initial confidence scores of the respective polysyllabic commands according to the determination of whether the speech signal corresponds to the monosyllabic form or the polysyllabic form;
• and
• a weight value applier calculating final confidence scores of the respective monosyllabic commands by applying the first weight value to the initial confidence scores of the respective monosyllabic commands and calculating final confidence scores of the respective polysyllabic commands by applying the second weight value to the initial confidence scores of the respective polysyllabic commands.

A recognition method comprising:

• detecting a speech segment having a start point and an end point by analyzing a speech signal within the speech segment;
• determining whether the speech signal corresponds to a monosyllabic form or a polysyllabic form;
• extracting a feature vector by analyzing the speech signal;
• selecting a monosyllabic command context including a plurality of monosyllabic commands and recognizing at least one command based on the extracted feature vector using the monosyllabic command context, when the speech signal corresponds to the monosyllabic form;
• and
• selecting a polysyllabic command context including a plurality of polysyllabic commands and recognizing at least one command based on the extracted feature vector using the polysyllabic command context, when the speech signal corresponds to the polysyllabic form.

A speech recognition method comprising:

• detecting a speech segment having a start point and an end point by analyzing a speech signal within the speech segment;
• determining whether the speech signal corresponds to a monosyllabic form or a polysyllabic form;
• extracting a feature vector by analyzing the speech signal;
• calculating initial confidence scores of respective monosyllabic commands included in a monosyllabic command context and initial confidence scores of respective polysyllabic commands included in a polysyllabic command context based on the extracted feature vector;
• setting a first weight value which is to be applied to the initial confidence scores of the respective monosyllabic commands and a second weight value which is to be applied to the initial confidence scores of the respective polysyllabic commands according to the determination of whether the speech signal corresponds to the monosyllabic form or the polysyllabic form;
• and
• calculating final confidence scores of the respective monosyllabic commands by applying the first weight value to the initial confidence scores of the respective monosyllabic commands and final confidence scores of the respective polysyllabic commands by applying the second weight value to the initial confidence scores of the respective polysyllabic commands.