Speech recognition apparatus using neural network and fuzzy logic

A recognition apparatus, comprising:

• input means for inputting an input signal;
• a plurality of processors each of which comprise a neural network for extracting at least one of a plurality of specific features from said input signal by a learning process;
• a plurality of first fuzzy logic means each for determining a specific feature having the greatest certainty of the specific features extracted by a respective one of said plurality of processors by processing related membership functions, and each for outputting said one specific feature having the greatest certainty as a partial recognition result;
• and
• second fuzzy logic means for determining the one of the partial recognition results outputted by said plurality of first fuzzy logic means which has the greatest certainty by processing related membership functions, and for outputting said one partial recognition result having the greatest certainty as a recognition result.

A recognition apparatus, comprising:

• input means for inputting an input signal;
• a first neural network for extracting at least one of a plurality of specific features from said input signal by a learning process;
• and
• a second neural network for extracting a recognition result from at least one of the extracted specific features from said first neural network by a learning process.

A recognition apparatus, comprising:

• input means for inputting an input signal;
• a first neural network for extracting at least one of a plurality of specific features from said input signal by a learning process;
• a second neural network for extracting a partial recognition result from at least one of the extracted specific features from said first neural network by a learning process;
• and
• a third neural network for extracting a recognition result from the extracted partial recognition result from said second neural network by a learning process.

A recognition apparatus, comprising:

• input means for inputting an input signal;
• a plurality of first neural networks each for extracting at least one of a plurality of specific features from said input signal by a learning process;
• a plurality of second neural networks each for extracting a partial recognition result from at least one of the extracted specific features from said plurality of first neural networks by a learning process;
• and
• fuzzy logic means for determining the one of the extracted partial recognition results of said plurality of second neural networks which has the greatest certainty by processing related membership functions, and for outputting said one partial recognition result having the greatest certainty as a recognition result.