Speech recognition system

Claim 2

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for calculating an asymptotic-processing formula based on the first distances, and for judging a word representing the sound based on a result of a calculation of said asymptotic-processing formula.

Claim 3

The system as recited in claim 1, further comprising: a first distance memory for storing the first distances; and a distance extracting unit for reading a predetermined number of the first distances from said first distance memory using a block transfer mode, and outputting the predetermined number of the first distances to said judging unit.

Claim 4

The system according to claim 1, wherein said plurality of frames are stored in said input pattern memory as a time block.

Claim 5

The system according to claim 1, wherein said distance calculating unit extracts features of the analyzed data when the input pattern memory stores a predetermined number of the features, said predetermined number being a time block.

Claim 6

The system according to claim 4, wherein said distance calculating unit extracts the features simultaneously when the input pattern memory stores the features.

Claim 7

The system according to claim 5, wherein said distance calculating unit accesses the input pattern memory only after the input pattern memory stores a last feature of the predetermined number of the features.

Claim 8

The system according to claim 1, wherein a number of accesses required by said distance calculating unit to access said standard pattern memory is (I×K)/(IL), where I is an integer and is a number of frames of said plurality of frames, K is an integer representing a total number of subword standard patterns stored in said standard pattern memory, and IL is a predetermined number of features of the plurality of frames.

Claim 9

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances; a first distance memory for storing the first distances; a distance extracting unit for reading a predetermined number of the first distances from said first distance memory using a block transfer mode, and outputting the predetermined number of the first distances to said judging unit; and a dictionary memory, coupled to said distance extracting unit, for storing words, wherein said judging unit comprises: a second distance memory for receiving the predetermined number of the first distances from said distance extracting unit and storing the predetermined number of the first distances; an asymptotic-processing formula calculating unit for calculating an asymptotic-processing formula based on the first distances stored in said second distance memory and said words stored in said dictionary memory; and a result determining unit for judging the word closest associated with the sound, based on a result of a calculation of said asymptotic-processing formula.

Claim 10

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances; a first distance memory for storing the first distances; a distance extracting unit for reading a predetermined number of the first distances from said first distance memory using a block transfer mode, and outputting the predetermined number of the first distances to said judging unit; and a dictionary memory, coupled to said distance extracting unit, for storing words, wherein said judging unit comprises: a second distance memory for receiving the predetermined number of the first distances from said distance extracting unit and storing the predetermined number of the first distances; an asymptotic-processing formula calculating unit for calculating an asymptotic-processing formula based on the first distances stored in said second distance memory and said words stored in said dictionary memory; and a result determining unit for judging the word closest associated with the sound, based on a result of a calculation of said asymptotic-processing formula.

Claim 11

The system as recited in claim 8, wherein said input pattern memory and said second distance memory each output data faster than each of said standard pattern memory, said first distance memory, and said dictionary memory.

Claim 12

The system as recited in claim 9, wherein said input pattern memory and said second distance memory comprise memories internal to a microprocessor.

Claim 13

The system as recited in claim 10, wherein said standard pattern memory, said first distance memory, and said dictionary memory comprise memories external to said microprocessor.

Claim 14

The system as recited in claim 11, wherein said second distance memory comprises a cache memory.

Claim 15

A speech recognition system for use with a standard pattern memory and a first distance memory comprising: a sound receiving unit for receiving a sound, converting the sound to a plurality of first data, calculating first distances between each one of the first data and one of second data stored in the standard pattern memory, and storing the first distances to the first distance memory; a distance extracting unit for reading the predetermined number of the first distances stored in said first distance memory by using a block transfer mode; and a judging unit for calculating an asymptotic-processing formula based on the first distances read by said distance extracting unit, and for judging a word representing the sound based on a result of a calculation of said asymptotic-processing formula.

Claim 16

A speech recognition system for use with a standard pattern memory and a first distance memory comprising: a sound receiving unit for receiving a sound, converting the sound to a plurality of first data, calculating first distances between each one of the first data and one of second data stored in the standard pattern memory, and storing the first distances to the first distance memory; a distance extracting unit for reading the predetermined number of the first distances stored in said first distance memory by using a block transfer mode; and a judging unit for calculating an asymptotic-processing formula based on the first distances read by said distance extracting unit, and for judging a word representing the sound based on a result of a calculation of said asymptotic-processing formula.

Claim 17

The system as recited in claim 13, further comprising an input pattern memory coupled to said sound receiving unit, wherein said sound receiving unit comprises: a speech input unit for sampling the sound; an analyzing unit for extracting the sound from said speech input unit, sequentially dividing the sound into a plurality of frames, converting each of the frames sequentially to the first data, and sequentially storing the first data to said input pattern memory; and a distance calculating unit for reading a predetermined number of the first data from said input pattern memory, reading one of the second data from said standard pattern memory, and calculating first distances between each of the predetermined number of the first data and the one of the second data.

Claim 18

A speech recognition system for use with a standard pattern memory and a first distance memory comprising: a sound receiving unit for receiving a sound, converting the sound to a plurality of first data, calculating first distances between each one of the first data and one of second data stored in the standard pattern memory, and storing the first distances to the first distance memory; a distance extracting unit for reading the predetermined number of the first distances stored in said first distance memory by using a block transfer mode; and a judging unit for judging a word representing the sound based on the first distances read by said distance extracting unit; an input pattern memory coupled to said sound receiving unit, wherein said sound receiving unit comprises: a speech input unit for sampling the sound; an analyzing unit for extracting the sound from said speech input unit, sequentially dividing the sound into a plurality of frames, converting each of the frames sequentially to the first data, and sequentially storing the first data to said input pattern memory; and a distance calculating unit for reading a predetermined number of the first data from said input pattern memory, reading one of the second data from said standard pattern memory, and calculating first distances between each of the predetermined number of the first data and the one of the second data; and a dictionary memory coupled to said distance extracting unit, wherein said judging unit comprises: a second distance memory for receiving the predetermined number of the first distances from said distance extracting unit and storing the predetermined number of the first distances; an asymptotic-processing formula calculating unit for calculating an asymptotic-processing formula based on the first distances stored in said second distance memory and words stored in a dictionary memory; and a result determining unit for judging the word associated with the sound based on a result of a calculation of said asymptotic-processing formula.

Claim 19

A speech recognition system for use with a standard pattern memory and a first distance memory comprising: a sound receiving unit for receiving a sound, converting the sound to a plurality of first data, calculating first distances between each one of the first data and one of second data stored in the standard pattern memory, and storing the first distances to the first distance memory; a distance extracting unit for reading the predetermined number of the first distances stored in said first distance memory by using a block transfer mode; and a judging unit for judging a word representing the sound based on the first distances read by said distance extracting unit; an input pattern memory coupled to said sound receiving unit, wherein said sound receiving unit comprises: a speech input unit for sampling the sound; an analyzing unit for extracting the sound from said speech input unit, sequentially dividing the sound into a plurality of frames, converting each of the frames sequentially to the first data, and sequentially storing the first data to said input pattern memory; and a distance calculating unit for reading a predetermined number of the first data from said input pattern memory, reading one of the second data from said standard pattern memory, and calculating first distances between each of the predetermined number of the first data and the one of the second data; and a dictionary memory coupled to said distance extracting unit, wherein said judging unit comprises: a second distance memory for receiving the predetermined number of the first distances from said distance extracting unit and storing the predetermined number of the first distances; an asymptotic-processing formula calculating unit for calculating an asymptotic-processing formula based on the first distances stored in said second distance memory and words stored in a dictionary memory; and a result determining unit for judging the word associated with the sound based on a result of a calculation of said asymptotic-processing formula.

Claim 20

The system as recited in claim 15, wherein said input pattern memory and said second distance memory each output data faster than said standard pattern memory, said first distance memory, and said dictionary memory.

Claim 21

The system as recited in claim 16, wherein said input pattern memory and said second distance memory comprise memories internal to a microprocessor.

Claim 22

The system as recited in claim 17, wherein said standard pattern memory, said first distance memory, and said dictionary memory comprise memories external to said microprocessor.

Claim 23

The system as recited in claim 18, wherein said second distance memory comprises a cache memory.

Claim 24

A speech recognition system comprising: a subword standard pattern memory for previously storing standard patterns formed in a unit of a subword; an analyzing unit for analyzing inputted speech in a unit of a frame to thereby obtain input patterns corresponding to time sequences of features of the inputted speech; an input pattern memory for storing said input patterns obtained in said analyzing unit; a distance calculating unit for calculating a distance between said input pattern and said standard pattern in each frame in a unit of a time block while said input patterns of a plurality of frames are combined together to form a single time block; a first distance memory for storing the distance value in said unit of said time block calculated in said distance calculating unit; a second distance memory for reading/writing data at a speed higher than that of said first distance memory, for accepting a transfer of the distance value in said unit of said time block calculated in said distance calculating unit to store the transferred distance value; a dictionary memory for storing information related to a word to be recognized; an accumulation value memory for storing an accumulation value of the distances; an asymptotic-processing formula calculating unit for calculating a new accumulation value of the distances in said unit of said time block based on information related to said word to be recognized stored in said dictionary memory, said distance value stored in the second distance memory, and said accumulation value stored in the accumulation value memory; and a recognition result determining unit for determining a recognition result based on the accumulation value of the distance stored in said accumulation value memory.

Claim 25

A speech recognition system comprising: a subword standard pattern memory for previously storing standard patterns formed in a unit of a subword; an analyzing unit for analyzing inputted speech in a unit of a frame to thereby obtain input patterns corresponding to time sequences of features of the inputted speech; an input pattern memory for storing said input patterns obtained in said analyzing unit; a distance calculating unit for calculating a distance between said input pattern and said standard pattern in each frame in a unit of a time block while said input patterns of a plurality of frames are combined together to form a single time block; a first distance memory for storing the distance value in said unit of said time block calculated in said distance calculating unit; a second distance memory for reading/writing data at a speed higher than that of said first distance memory, for accepting a transfer of the distance value in said unit of said time block calculated in said distance calculating unit to store the transferred distance value; a dictionary memory for storing information related to a word to be recognized; an accumulation value memory for storing an accumulation value of the distances; an asymptotic-processing formula calculating unit for calculating a new accumulation value of the distances in said unit of said time block based on information related to said word to be recognized stored in said dictionary memory, said distance value stored in the second distance memory, and said accumulation value stored in the accumulation value memory; and a recognition result determining unit for determining a recognition result based on the accumulation value of the distance stored in said accumulation value memory.

Claim 26

The system as recited in claim 20, further comprising a distance extracting unit for extracting a distance value in said unit of said time block from said first distance memory based on said information held in said dictionary memory, and block-transferring the extracted distance value in said unit of said time block to said second distance memory.

Claim 27

The system as recited in claim 21, wherein said second distance memory comprises a cache memory.

Claim 28

The system as recited in claim 20, wherein both said input pattern memory and said second distance memory comprise memories operable at a speed higher than that of said first distance memory.

Claim 29

The system as recited in claim 23, wherein said second distance memory comprises a cache memory.

Claim 30

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances, wherein said plurality of frames are stored in said input pattern memory as a time block, and wherein said distance calculating unit extracts the features of the analyzed data when the input pattern memory stores a predetermined number of the features, said predetermined number representing said time block.

Claim 31

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances, wherein said plurality of frames are stored in said input pattern memory as a time block, and wherein said distance calculating unit extracts the features of the analyzed data when the input pattern memory stores a predetermined number of the features, said predetermined number representing said time block.

Claim 32

The system according to claim 25, wherein said distance calculating unit extracts the features simultaneously when the input pattern memory stores the features.

Claim 33

The system according to claim 26, wherein said distance calculating unit accesses the input pattern memory only after the input pattern memory stores a last feature of the predetermined number of the features.

Claim 34

The system according to claim 25, wherein a number of accesses required by said distance calculating unit to access said standard pattern memory is (I×K)/(IL), where I is an integer and is a number of frames of said plurality of frames, K is an integer representing a total number of subword standard patterns stored in said standard pattern memory, and IL is a predetermined number of features of the plurality of frames.

Claim 35

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances, wherein said distance calculating unit extracts the features of the analyzed data when the input pattern memory stores a predetermined number of the features, said predetermined number representing a time block.

Claim 36

A speech recognition system, comprising: an analyzing unit for receiving a sound, sequentially dividing the sound into a plurality of frames, and converting each of the frames sequentially to first data; an input pattern memory for sequentially storing the first data; a distance calculating unit for reading a predetermined number of the first data from said input pattern memory; a standard pattern memory for storing second data, said distance calculating unit for reading one of said second data, and for calculating first distances between each of the predetermined number of the first data and the one of the second data; and a judging unit for judging a word representing the sound based on the first distances, wherein said distance calculating unit extracts the features of the analyzed data when the input pattern memory stores a predetermined number of the features, said predetermined number representing a time block.

Claim 37

The system according to claim 29, wherein said distance calculating unit extracts the features simultaneously when the input pattern memory stores the features.

Claim 38

The system according to claim 30, wherein said distance calculating unit accesses the input pattern memory only after the input pattern memory stores a last feature of the predetermined number of the features.

Claim 39

The system according to claim 29, wherein a number of accesses required by said distance calculating unit to access said standard pattern memory is (I×K)/(IL), where I is an integer and is a number of frames of said plurality of frames, K is an integer representing a total number of subword standard patterns stored in said standard pattern memory, and IL is a predetermined number of features of the plurality of frames.