Method and apparatus for speech recognition using both low-order and high-order parameter analyzation

A speech recognition apparatus comprising:

• low-order parameter analyzation means for deriving a low-order parameter time series from input speech and outputting speech energy;
• start and end point candidate detection means for detecting the candidates of the start and end points of the input speech according to the relation between the level of the speech energy output from said low-order parameter analyzation means and a threshold value;
• high-order parameter analyzation means for analyzing said input speech and intermittently deriving high order parameters at preset time intervals along a time axis within a range of the input speech between the candidates of the start and end point detected by said start and end point candidate detection means, on the basis of the speech energy output from said low-order parameter analyzation means;
• start and end point decision means for checking the relation between the speech energy level and the threshold value and determining the candidates of the start and end points as the start and end points;
• and
• identifying means for identifying the input speech based on the parameters derived by said high-order parameter analyzation means in response to the determining of the start and end points.

A speech recognition method comprising the steps of:

• a) analyzing input speech by a predetermined number of arithmetic operations and deriving a low-order parameter time series of the input speech by using a personal computer;
• b) detecting start and end points of the input speech from the parameter time series derived by said low-order parameter analyzation;
• c) analyzing said input speech by a number of arithmetic operations larger than the predetermined number of arithmetic operations for deriving the low-order parameter time series, to intermittently derive high order parameters at preset time intervals along a time axis within a range of the input speech, extending from the start point to the end point by using said personal computer after the start and end points are detected;
• and
• d) matching feature parameters corresponding to the high order parameters derived by said high-order parameter analyzation with reference parameters previously registered in said personal computer so as to enable recognition of the input speech according to a result of a collation process.

A speech recognition method comprising the steps of:

• a) analyzing input speech by a predetermined number of arithmetic operations for deriving a low-order parameter time series from the input speech and outputting low-order speech energy by using a personal computer;
• b) detecting candidates of start and end points of the input speech according to a relation between a level of the low-order speech energy and a threshold value;
• c) analyzing the input speech by a number of arithmetic operations larger than the predetermined number of arithmetic operations for deriving the low-order time series, to intermittently derive high order parameters at preset time intervals along a time axis within a range of the input speech between the detected candidates of the start and end points by using said personal computer;
• d) checking the relation between the speech energy level and the threshold value and determining the start and end point candidates as the start and end points;
• and
• e) identifying the input speech based on the parameters derived by step c) in response to the determination of the start and end points.