Speech recognition activation and deactivation method

Claim 2

In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and means responsive to said acoustic parameters for generating likelihood costs between said acoustic parameters and said speech template patterns, and for processing said likelihood costs for determining the speech units in said speech input signal, a method of template matching and cost processing for recognizing the correspondence of said speech input signal and said template patterns, said method comprising the steps of characterizing the allowable possible sequences of speech units as a grammar graph, said graph having a plurality of grammar nodes connected by a plurality of connecting arcs, each said arc having associated therewith at least one word, each word having at least one kernel, and each kernel having one template pattern, deactivating kernels of each said word having a plurality of kernels when a minimum cumulative score associated therewith exceeds a deactivation threshold, kernels which have not been deactivated being called active kernels, generating likelihood costs representing the similarity of said acoustic parameters and ones of said active kernels, determining, at each frame time, cumulative scores associated with said nodes, generating a speech recognition decision, and determining from said cumulative scores the identity of the speech units in said speech input signal.

Claim 3

In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and means responsive to said acoustic parameters for generating likelihood costs between said acoustic parameters and said speech template patterns, and for processing said likelihood costs for determining the speech units in said speech input signal, a method of template matching and cost processing for recognizing the correspondence of said speech input signal and said template patterns, said method comprising the steps of characterizing the allowable possible sequences of speech units as a grammar graph, said graph having a plurality of grammar nodes connected by a plurality of connecting arcs, each said arc having associated therewith at least one word, each word having at least one kernel associated therewith and each kernel having one template pattern, activating kernels of each said word having a plurality of kernels when a cumulative score associated with a previous kernel exceeds an activation threshold, kernels which have not been activated being called inactive kernels, generating likelihood costs representing the similarity of said acoustic parameters and ones of said active kernels, determining, at each frame time, cumulative scores associated with said nodes, generating a speech recognition decision, and determining from said cumulative scores the identity of the speech units in said speech input signal.

Claim 4

In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and means responsive to said acoustic parameters for generating likelihood costs between said acoustic parameters and said speech template patterns, and for processing said likelihood costs for determining the speech units in said speech input signal, a method of template matching and cost processing for recognizing the correspondence of said speech input signal and said template patterns, said method comprising the steps of characterizing the allowable possible sequences of speech units as a grammar graph, said graph having a plurality of grammar nodes connected by a plurality of connecting arcs, each said arc having associated therewith at least one word, each word having at least one kernel associated therewith and each kernel having one template pattern, activating kernels of each said word having a plurality of kernels when a cumulative score associated with a previous kernel exceeds an activation threshold, kernels which have not been activated being called inactive kernels, generating likelihood costs representing the similarity of said acoustic parameters and ones of said active kernels, determining, at each frame time, cumulative scores associated with said nodes, generating a speech recognition decision, and determining from said cumulative scores the identity of the speech units in said speech input signal.

Claim 5

In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and means responsive to said acoustic parameters for generating likelihood costs between said acoustic parameters and said speech template patterns, and for processing said likelihood costs for determining the speech units in said speech input signal, a method of template matching and cost processing for recognizing the correspondence of said speech input signal and said template patterns, said method comprising the steps of characterizing the allowable possible sequences of speech units as a grammar graph, said graph having a plurality of grammar nodes connected by a plurality of connecting arcs, each said arc having associated therewith at least one word, each word having associated therewith at least one kernel, and each kernel having one template pattern, deactivating kernels of each said word having a plurality of kernels when a cumulative score associated therewith exceeds a deactivation threshold, kernels which have not been deactivated being called active kernels, activating kernels of each said word having a plurality of kernels when a cumulative score associated with a previous kernel exceeds an activation threshold, kernels which have not been activated being called inactive kernels, generating likelihood costs representing the similarity of said acoustic parameters and ones of said active kernels, determining, at each frame time, cumulative scores associated with said nodes, generating a speech recognition decision, and determining from said cumulative scores the identity of the speech units in said speech input signal.

Claim 6

In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and means responsive to said acoustic parameters for generating likelihood costs between said acoustic parameters and said speech template patterns, and for processing said likelihood costs for determining the speech units in said speech input signal, a method of template matching and cost processing for recognizing the correspondence of said speech input signal and said template patterns, said method comprising the steps of characterizing the allowable possible sequences of speech units as a grammar graph, said graph having a plurality of grammar nodes connected by a plurality of connecting arcs, each said arc having associated therewith at least one word, each word having associated therewith at least one kernel, and each kernel having one template pattern, deactivating kernels of each said word having a plurality of kernels when a cumulative score associated therewith exceeds a deactivation threshold, kernels which have not been deactivated being called active kernels, activating kernels of each said word having a plurality of kernels when a cumulative score associated with a previous kernel exceeds an activation threshold, kernels which have not been activated being called inactive kernels, generating likelihood costs representing the similarity of said acoustic parameters and ones of said active kernels, determining, at each frame time, cumulative scores associated with said nodes, generating a speech recognition decision, and determining from said cumulative scores the identity of the speech units in said speech input signal.