EP0421744A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Speech recognition method and apparatus for use therein
Assignee: Canon Inc
•
Filing: 1990-10-03
•
Granted: —
•
Expires: 2010-10-03
Abstract
Abstract A speech recognition method and apparatus includes a microphone for inputting a spoken word, a memory for storing a plurality of data representing a standard pattern of a spoken word composed of an average vector and a variance/covariance matrix, a device for shifting the standard pattern, a calculator for calculating a plurality of Maharanobis distances between the data representing a time series of parameters of the spoken word input by the microphone and a plurality of data representing the shifting standard pattern, a detector for detecting data representing the minimum Maharanobis distance calculated by the calculator, and a recognizing device for recognizing whether the input spoken word is the word represented by the standard pattern based on the data representing the minimum Maharanobis distance detected by the detector. The recognizing device calculates a plurality of dynamic programming values of the inputted spoken word over time while the standard pattern is shifted. The recognizing device also determines whether one of the plurality of the dynamic programming values is below a predetermined threshold value, and then recognizes the inputted spoken word as the word represented by the standard pattern when the recognizing device determines that one of the plurality of dynamic programming values is below the predetermined threshold value.
Inventors
- 👤 Tetsuo Kosaka
- 👤 Atsushi Sakurai
- 👤 Junichi Tamura
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-06 00:19
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/12; G10L15/10
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
162 words
A speech recognition apparatus, comprising: input means for inputting a spoken word; memory means for storing data representing a standard pattern of a spoken word composed of an average vector and a variance/covariance matrix; shift means for shifting the data representing the standard pattern; calculation means for calculating a plurality of Maharanobis distances between data representing a time series of parameters of the spoken word input by said input means and data representing the shifting standard pattern stored in said memory means; retention means for retaining data representing the plurality of Maharanobis distances together with the data representing the standard patterns used by said calculation means for the calculations; detection means for detecting data representing the minimum Maharanobis distance retained in said retention means; and recognition means for recognizing whether the input spoken word is the word represented by the standard pattern based on the data representing the minimum Maharanobis distance retained by said retention means and detected by said detection means.
Claim 2
Dependent Claim
162 words
A speech recognition apparatus, comprising: input means for inputting a spoken word; memory means for storing data representing a standard pattern of a spoken word composed of an average vector and a variance/covariance matrix; shift means for shifting the data representing the standard pattern; calculation means for calculating a plurality of Maharanobis distances between data representing a time series of parameters of the spoken word input by said input means and data representing the shifting standard pattern stored in said memory means; retention means for retaining data representing the plurality of Maharanobis distances together with the data representing the standard patterns used by said calculation means for the calculations; detection means for detecting data representing the minimum Maharanobis distance retained in said retention means; and recognition means for recognizing whether the input spoken word is the word represented by the standard pattern based on the data representing the minimum Maharanobis distance retained by said retention means and detected by said detection means.
Claim 3
Dependent Claim
48 words
A speech recognition apparatus according to claim 1, wherein said calculation means calculates each of the plurality of distances by using a measure of the power of the inputted spoken word, the average vector of the inputted spoken word, and the variance/covariance matrix of the inputted spoken word.
Claim 4
Dependent Claim
30 words
A speech recognition apparatus according to claim 1, further comprising means for converting the input spoken word into data representing a time series of parameters of the input spoken word.
Claim 5
Dependent Claim
93 words
A speech recognition apparatus according to claim 1, wherein said recognition means comprises: calculation means for calculating a plurality of dynamic programming values of the inputted spoken word over time while said shift means shifts the standard pattern; determining means for determining whether one of said plurality of dynamic programming values is below a predetermined threshold value; and means for recognizing the inputted spoken word as the word represented by the standard pattern when said determining means determines that one of said plurality of dynamic programming values is below the predetermined threshold value.
Claim 6
Dependent Claim
17 words
A speech recognition apparatus according to claim 4, further comprising means for setting the predetermined threshold value.
Claim 7
Independent Claim
90 words
A method for recognizing speech comprising the steps of: converting a spoken word into data representing a time series of parameters; calculating a plurality of Maharanobis distances between the data representing the time series of parameters and data representing a shifting standard pattern of a spoken word comprising an average vector and a variance/covariance matrix; detecting the minimum Maharanobis distance; and recognizing whether the spoken word converted in said converting step is the word represented by the standard pattern based on the minimum Maharanobis distance detected in said detecting step.
Claim 8
Independent Claim
90 words
A method for recognizing speech comprising the steps of: converting a spoken word into data representing a time series of parameters; calculating a plurality of Maharanobis distances between the data representing the time series of parameters and data representing a shifting standard pattern of a spoken word comprising an average vector and a variance/covariance matrix; detecting the minimum Maharanobis distance; and recognizing whether the spoken word converted in said converting step is the word represented by the standard pattern based on the minimum Maharanobis distance detected in said detecting step.
Claim 9
Dependent Claim
94 words
A method according to claim 6, wherein said recognizing step comprises the steps of: calculating a plurality of dynamic processing values over time of the spoken word converted in said converting step using the shifting standard pattern; determining whether one of the plurality of dynamic programming values is below a predetermined threshold value; and recognizing the spoken word converted in said converting step as the word represented by the standard pattern when said one of the plurality of dynamic programming values is determined in said determining step to be below the predetermined threshold value.
Claim 10
Dependent Claim
16 words
A method according to claim 7, further comprising the step of setting the predetermined threshold value.
Claim 11
Dependent Claim
62 words
A method according to claim 6, wherein said calculation step comprises the step of calculating each of the plurality of distances by using a measure of the power of the spoken word converted in said converting step, the average vector of the spoken word converted in said converting step, and the variance/covariance matrix of the spoken word converted in the converting step.
Claim 12
Dependent Claim
55 words
A speech recognition apparatus or method in which a standard pattern for a word to be recognised is compared with an audio input at a plurality of time offsets including time offsets for which the part of the audio input with which the standard pattern is compared has not previously been identified as including speech.
Claim 13
Dependent Claim
55 words
A speech recognition apparatus or method in which a standard pattern for a word to be recognised is compared with an audio input at a plurality of time offsets including time offsets for which the part of the audio input with which the standard pattern is compared has not previously been identified as including speech.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.