System and method for automated speech recognition

A system for automatically recognizing speech in a noisy environment, the system comprising:

• an acquisition module to obtain an auditory signal from the noisy environment;
• and
• a controller in operative communication with the acquisition module, the controller being configured to:
• obtain an auditory signal from the acquisition module,
• process the auditory signal into a plurality of frequency components,
• process the plurality of frequency components using a plurality of artificial neurons, each one of the plurality of artificial neurons producing a feature detector response that corresponds to a simulated auditory neuron response to speech,
• generate a spike for each instance in which a feature detector response identifies a characteristic spectro-temporal pattern to produce a sparse test spike pattern for the auditory signal, wherein the spike represents speech in terms of an appearance of the characteristic spectro-temporal pattern contained in a segment of speech as a function of time,
• weight the feature detector response corresponding to the spike such that each feature detector response discriminates spectro-temporal features at particular times in a speech element to spectro-temporal features in other speech elements;
• select a group of the plurality of artificial neurons such that a sequential production of feature detector responses is generated by the group for a given speech element, wherein each feature detector response occurs once during the given speech element, creating the sparse test spike pattern representative of the given speech element,
• compare the sparse test spike pattern to a plurality of predetermined sparse spike patterns corresponding to a plurality of speech elements,
• determine that the auditory signal includes one of the plurality of speech elements based on the comparing,
• generate a plurality of neuron labels corresponding to the plurality of speech elements;
• and
• output a plurality of neuron labels to a device that implements the plurality of neuron labels in detecting speech elements in a noisy signal.

A method implemented by a computing device for automatically recognizing continuous speech in a noisy environment, the method comprising:

• obtaining an auditory signal comprising a plurality of syllables of speech from the noisy environment;
• identifying a syllable within the auditory signal, the syllable comprising a portion of the auditory signal;
• processing the syllable into a plurality of frequency components;
• processing the plurality of frequency components using a plurality of artificial neurons, each one of the artificial neurons producing a feature detector response;
• generating a spike for each instance in which a feature detector response identifies a characteristic spectro-temporal pattern to produce a sparse test spike pattern for the auditory signal, wherein the spike represents speech in terms of an appearance of the characteristic spectro-temporal pattern contained in the syllable as a function of time;
• weighting the feature detector response corresponding to the spike such that each feature detector response discriminates spectro-temporal features at particular times in a speech element to spectro-temporal features in other speech elements;
• selecting a group of the plurality of artificial neurons such that a sequential production of feature detector responses is generated by the group for a given speech element, wherein each feature detector response occurs once during the given speech element, creating the sparse test spike pattern representative of the given speech element;
• comparing the sparse test spike pattern to a plurality of predetermined spike patterns corresponding to a plurality of speech elements;
• determining that the syllable corresponds to one of the plurality of speech elements based on the comparing;
• generating a plurality of neuron labels corresponding to the plurality of speech elements;
• and
• outputting a plurality of neuron labels to a device that implements the plurality of neuron labels in detecting speech elements in a noisy signal.