Model substitution for efficient development of a solution
AssigneeGroq, Inc.
InventorsGurshaant Singh Malik, Tim Sears
Life of this patent
Estimating the term…
Abstract
A neural network architecture for natural language processing is provided. The neural network architecture comprises: a speech-to-text encoder configured to encode an input speech signal; a Bifrost speech recognizable engine configured for processing the encoded speech signal to generate a speech recognized signal corresponding to the input speech signal; and a decoder configured to decode the speech recognized signal and to generate the output sequence.
Claims20
A neural network architecture for natural language processing comprising:
- a machine learning model configured for processing a speech signal to generate an intermediate speech recognition output corresponding to the speech signal, the machine learning model comprising a plurality of parallel dilated convolutional components, wherein each respective dilated convolutional component of the plurality of parallel dilated convolutional components implements a filter having gaps between its values, and wherein at least one respective dilated convolutional component of the plurality of parallel dilated convolutional components has a distinct dilation parameter that is different from a plurality of dilation parameters of other dilated convolutional components of the plurality of parallel dilated convolutional components;
- and
- a decoder configured to generate an output sequence based at least in part on the intermediate speech recognition output.
A non-transitory computer readable storage medium that stores computer program instructions, the computer program instructions, when executed by a computer processor, cause the computer processor to:
- provide an input speech signal to a machine learning model configured for processing the input speech signal to generate an intermediate speech recognition output corresponding to the input speech signal, the machine learning model comprising a plurality of parallel dilated convolutional components, wherein each respective dilated convolutional component of the plurality of parallel dilated convolutional components implements a filter having gaps between its values, and wherein at least one respective dilated convolutional component of the plurality of parallel dilated convolutional components has a distinct dilation parameter that is different from a plurality of dilation parameters of other dilated convolutional components of the plurality of parallel dilated convolutional components;
- and
- provide the intermediate speech recognition output to a decoder configured to generate an output sequence based at least in part on the intermediate speech recognition output.
A computing system comprising one or more processors and one or more non-transitory computer readable storage media that store instructions that, when executed by the one or more processors, cause the one or more processors to:
- provide an input speech signal to a machine learning model configured for processing the input speech signal to generate an intermediate speech recognition output corresponding to the input speech signal, the machine learning model comprising a plurality of parallel dilated convolutional components, wherein each respective dilated convolutional component of the plurality of parallel dilated convolutional components implements a filter having gaps between its values, and wherein at least one respective dilated convolutional component of the plurality of parallel dilated convolutional components has a distinct dilation parameter that is different from a plurality of dilation parameters of other dilated convolutional components of the plurality of parallel dilated convolutional components;
- and
- provide the intermediate speech recognition output to a decoder configured to generate an output sequence based at least in part on the intermediate speech recognition output.
Dependencies are inferred from the claim text.