Continuous speech recognition method for improving false alarm rates

Claim 2

In a speech analysis system for recognizing at least one predetermined keyword in an audio signal, each said keyword being characterized by a template having at least one target pattern and each target pattern representing at least one short-term power spectrum, an analysis method comprising the steps of forming a sequence of electrical signals repesenting a sequence of selected patterns, identifying electrical signals representing a candidate keyword when said sequence of selected patterns corresponds respectively to the sequence of target patterns of said keyword template, normalizing electrical signals representing the time duration spacings between selected patterns corresponding to said candidate word, and applying a prosodic test to said normalized time duration spacings, wherein said normalized time duration spacings for a candidate word must meet the timing criteria imposed by said prosodic test before said candidate word is accepted as a recognized keyboard.

Claim 3

The method of claim 1 wherein said applying step comprises the step of applying fixed, predetermined spacing limits to each normalized spacing, whereby said normalized spacings must fall within said fixed limits before said candidate word is accepted.

Claim 4

The method of claim 1 wherein said applying step comprises the steps of applying a likelihood statistic function to said normalized spacings and accepting said candidate word if the likelihood statistic exceeds a predetermined minimum threshold.

Claim 5

The method of claim 1 further comprising the steps of applying a likelihood ratio test to said sequence of selected patterns corresponding to said candidate keyword to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word if said accumulated figure of merit exceeds a predetermined minimum value.

Claim 6

The method of claim 4 wherein said applying step comprises the steps of determining a best value of the log likelihood statistic for each of said selected patterns relative to any of said target patterns, said best values being designated the best general scores, determining the best value of the log likelihood statistic for each of said selected patterns relative to those target patterns which are valid alternatives for the corresponding target patterns of the candidate keyword, said best values being designated the best target scores, and determining a figure of merit for each selected pattern by generating the arithmetic difference of the best general score and the best target score for the selected pattern.

Claim 7

In a speech analysis system for recognizing at least one predetermined keyboard in an audio signal, each said keyword being characterized by a template having at least one target pattern, and each target pattern representing at least one short-term power spectrum, an analysis method comprising the steps of forming a sequence of electrical signals representing a sequence of selected patterns, identifying electrical signals representing a candidate keyword when said sequence of selected patterns corresponds respectively to the sequence of target patterns of a said keyword template, applying a likelihood ratio test to the electrical signals representing said sequence of selected patterns corresponding to said candidate word to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word as a recognized keyword if said accumulated figure of merit exceeds a predetermined minimum value.

Claim 8

In a speech analysis system for recognizing at least one predetermined keyboard in an audio signal, each said keyword being characterized by a template having at least one target pattern, and each target pattern representing at least one short-term power spectrum, an analysis method comprising the steps of forming a sequence of electrical signals representing a sequence of selected patterns, identifying electrical signals representing a candidate keyword when said sequence of selected patterns corresponds respectively to the sequence of target patterns of a said keyword template, applying a likelihood ratio test to the electrical signals representing said sequence of selected patterns corresponding to said candidate word to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word as a recognized keyword if said accumulated figure of merit exceeds a predetermined minimum value.

Claim 9

The method of claim 6 wherein said applying step comprises the steps of determining a best value of the log likelihood statistic for each of said selected patterns relative to any of said target patterns, said best values being designated the best general scores, determining the best value of the log likelihood statistic for each of said selected patterns relative to those target patterns which are valid alternatives for the corresponding target patterns of the candidate keyword, said best values being designated the best target scores, and determining a figure of merit for each selected pattern by generating the arithmetic difference of the best general score and the best target score for the selected pattern.

Claim 10

In a speech analysis system in which an audio signal is spectrum analyzed for recognizing at least one predetermined keyword in a continuous audio signal, each said keyword being characterized by a template having at least one target pattern representing a plurality of short-term power spectra spaced apart in real time, an analysis method comprising the steps of repeatedly evaluating electrical signals representing a set of parameters determining a short-term power spectrum of said audio signal within each of a plurality of equal duration sampling intervals thereby to generate an uninterrupted time-ordered sequence of short-term audio power spectrum frames, repeatedly selecting from said sequence of frames, one first frame and at least one later occurring frame to form a multi-frame pattern, comparing each thus formed multi-frame pattern with each first target pattern of each keyword template, deciding whether each said multi-frame pattern corresponds to a first target pattern of a keyword template, for each multi-frame pattern which, according to said deciding step, corresponds to a said first target pattern of a potential candidate keyword, selecting later occurring short-term power spectra to form later occurring multi-frame patterns, deciding whether said later occurring multi-frame patterns correspond respectively to successive target patterns of said potential candidate keyword template, identifying electrical signals representing a candidate keyword template when said selected multi-frame patterns correspond respectively to the target patterns of a said keyword template, normalizing electrical signals representing the time duration spacings between multi-frame patterns corresponding to said candidate word, and applying a prosodic test to said normalized time duration spacings, wherein said normalized time duration spacings for a candidate word must meet the timing criteria imposed by said prosodic test before a said candidate word is accepted as a recognized keyword.

Claim 11

In a speech analysis system in which an audio signal is spectrum analyzed for recognizing at least one predetermined keyword in a continuous audio signal, each said keyword being characterized by a template having at least one target pattern representing a plurality of short-term power spectra spaced apart in real time, an analysis method comprising the steps of repeatedly evaluating electrical signals representing a set of parameters determining a short-term power spectrum of said audio signal within each of a plurality of equal duration sampling intervals thereby to generate an uninterrupted time-ordered sequence of short-term audio power spectrum frames, repeatedly selecting from said sequence of frames, one first frame and at least one later occurring frame to form a multi-frame pattern, comparing each thus formed multi-frame pattern with each first target pattern of each keyword template, deciding whether each said multi-frame pattern corresponds to a first target pattern of a keyword template, for each multi-frame pattern which, according to said deciding step, corresponds to a said first target pattern of a potential candidate keyword, selecting later occurring short-term power spectra to form later occurring multi-frame patterns, deciding whether said later occurring multi-frame patterns correspond respectively to successive target patterns of said potential candidate keyword template, identifying electrical signals representing a candidate keyword template when said selected multi-frame patterns correspond respectively to the target patterns of a said keyword template, normalizing electrical signals representing the time duration spacings between multi-frame patterns corresponding to said candidate word, and applying a prosodic test to said normalized time duration spacings, wherein said normalized time duration spacings for a candidate word must meet the timing criteria imposed by said prosodic test before a said candidate word is accepted as a recognized keyword.

Claim 12

The method of claim 8 wherein said applying step comprises the step of applying fixed, predetermined spacing limits to each normalized spacing, whereby said normalized spacings must fall within said fixed time limits before said candidate word is accepted.

Claim 13

The method of claim 8 wherein said applying step comprises the steps of applying a likelihood statistic function to said normalized spacings and accepting said candidate word if the likelihood statistic exceeds a predetermined minimum threshold.

Claim 14

The method of claim 8 further comprising the steps of applying a likelihood ratio test to said sequence of multi-frame patterns corresponding to a candidate word to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word if said accumulated figure of merit exceeds a predetermined minimum value.

Claim 15

The method of claim 11 wherein said applying step comprises the steps of determining a best value of the log likelihood statistic for each of said selected multi-frame patterns relative to any of said target patterns, said best values being designated the best general scores, determining the best value of the log likelihood statistic for each of said selected multi-frame patterns relative to those target patterns which are valid alternatives for the corresponding target patterns of the candidate keyword, said best values being designated the best target scores, and determining a figure of merit for each selected multi-frame pattern by generating the arithmetic difference of the corresponding best general score and the best target score for the selected pattern.

Claim 16

In a speech analysis system in which an audio signal is spectrum analyzed for recognizing at least one predetermined keyword in a continuous audio signal, each said keyword being characterized by a template having at least one target pattern, said target pattern having an ordered sequence and each target pattern representing a plurality of short term power spectra spaced apart in real time, an analysis method comprising the steps of repeatedly evaluating electrical signals representing a set of parameters determining a short-term power spectrum of said audio signal within each of a plurality of equal duration sampling intervals thereby to generate a continuous time-ordered sequence of short-term audio power spectrum frames, repeatedly selecting from said sequence of frames, one first frame and at least one later occurring frame to form a multi-frame pattern, comparing each thus formed multi-frame pattern with each first target pattern of each keyword template, deciding whether each said multi-frame pattern corresponds to said first target pattern of a keyword template, for each multi-frame pattern which, according to said deciding step, corresponds to a said first target pattern of a potential candidate keyword, selecting later occurring short-term power spectra to form later occurring multi-frame patterns, deciding whether said later occurring multi-frame patterns correspond respectively to successive target patterns of said potential candidate keyword template, identifying electrical signals representing a candidate keyword template when said selected multi-frame patterns correspond respectively to the target patterns of a said keyword template, applying a likelihood ratio test to electrical signals representing said sequence of multi-frame patterns corresponding to a candidate word to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word if said accumulated figure of merit exceeds a predetermined minumum value.

Claim 17

In a speech analysis system in which an audio signal is spectrum analyzed for recognizing at least one predetermined keyword in a continuous audio signal, each said keyword being characterized by a template having at least one target pattern, said target pattern having an ordered sequence and each target pattern representing a plurality of short term power spectra spaced apart in real time, an analysis method comprising the steps of repeatedly evaluating electrical signals representing a set of parameters determining a short-term power spectrum of said audio signal within each of a plurality of equal duration sampling intervals thereby to generate a continuous time-ordered sequence of short-term audio power spectrum frames, repeatedly selecting from said sequence of frames, one first frame and at least one later occurring frame to form a multi-frame pattern, comparing each thus formed multi-frame pattern with each first target pattern of each keyword template, deciding whether each said multi-frame pattern corresponds to said first target pattern of a keyword template, for each multi-frame pattern which, according to said deciding step, corresponds to a said first target pattern of a potential candidate keyword, selecting later occurring short-term power spectra to form later occurring multi-frame patterns, deciding whether said later occurring multi-frame patterns correspond respectively to successive target patterns of said potential candidate keyword template, identifying electrical signals representing a candidate keyword template when said selected multi-frame patterns correspond respectively to the target patterns of a said keyword template, applying a likelihood ratio test to electrical signals representing said sequence of multi-frame patterns corresponding to a candidate word to determine a figure of merit for each said pattern, accumulating said figures of merit for said patterns, and accepting said candidate word if said accumulated figure of merit exceeds a predetermined minumum value.

Claim 18

The method of claim 13 wherein said applying step comprises the steps of determining a best value of the log likelihood statistic for each of said selected multi-frame patterns relative to any of said target patterns, said best values being designated the best general scores, determining the best value of the log likelihood statistic for each of said selected multi-frame patterns relative to those target patterns which are valid alternatives for the corresponding target patterns of the candidate keyword, said best values being designated the best target scores, and determining a figure of merit for each selected multi-frame pattern by generating the arithmetic difference of the corresponding best general score and the best target score for the selected pattern.