US 2020/0286480 A1Published application

Brain-inspired spoken language understanding system, a device for implementing the system, and method of operation thereof

AssigneeMedyug Technology Private Limited

InventorsBaljit Singh, Praveen Prakash

Life of this patent

Estimating the term…

Loading patent…

Abstract

A brain-inspired spoken language understanding system, comprises: a first module that facilitates the conversion of a voice input into phoneme sequences; a buffer module that facilitates the storing of the phoneme sequences until they are clustered into a first storage as one or more meaningful thought representations; a second module that facilitates the monitoring of the reasoning, disambiguation, and prioritization of the system, in addition to controlling the system; a seventh module and an eighth module that facilitate the capturing of at least one non-verbal signal from a conversation, and create at least one time-synchronized non-verbal object, with the help of a third knowledge base; a third module that facilitates the conversion of the phoneme sequences and at least one time-synchronized non-verbal object into sequences of phonetics-based words; a fourth module that facilitates the conversion of the phonetics-based words into sequences of thought representations, with the help of a first knowledge base; a fifth module that facilitates the fine-tuning of the thought representations, with the help of the first knowledge base and a second knowledge base; and a sixth module that maintains the instance of conversation as thought representations in the form of one or more connected stories. A device for implementing the brain-inspired spoken language understanding system and its method of functioning are also disclosed.

A brain-inspired spoken language understanding system, comprising:

a first module that facilitates the conversion of a voice input received through one or more microphones into phoneme sequences, said first module comprising:
a first unit that converts the voice input received into spectrograms, and a second unit that converts the spectrograms into phonemes, with pause time between them, if applicable;
a buffer module that facilitates the storing of the phoneme sequences until they are clustered into a first storage as one or more meaningful thought representations;
a second module that facilitates the monitoring of the reasoning, disambiguation, and prioritization of the system, in addition to controlling the system;
a seventh module that comprises:
at least one sensor that captures at least one non-verbal signal from a conversation, and a pre-processing unit that facilitates the pre-processing of the at least one non-verbal signal received from the at least one sensor;
an eighth module that is configured to classify and transform the at least one non-verbal signal received from the seventh module into at least one symbol, and creates at least one time-synchronized non-verbal object by permuting and matching the at least one symbol against a third knowledge base, said third knowledge base providing a look-up for non-verbal symbols into known complex objects;
a ninth module that facilitates the transmission of the at least one object received from the eighth module, along with time-stamp, to a third module;
the third module that facilitates the conversion of the phoneme sequences and at least one time-synchronized non-verbal object into sequences of phonetics-based words;
the fourth module that facilitates the conversion of the phonetics-based words into sequences of thought representations, with the help of a first knowledge base;
a fifth module that facilitates the fine-tuning of the thought representations received from the fourth module by:
merging two or more thought representations, inferring new thought representations, correcting the thought representations, and/or filling one or more missing information in the thought representations, with the help of the first knowledge base and a second knowledge base;
and
a sixth module that maintains the instance of conversation as thought representations in the form of one or more connected stories as thought cloud, with said third module, said fourth module, and said fifth module interacting with each other, with the help of the second module, and with the system running the sequence possibilities throughout said third module, said fourth module, and said fifth module like a pipeline with feedback coming from forward modules to prior modules.

11 more elements

Facts

Application

16/810,358Filed 5 Mar 2020

Priority

IN 2019410084815 Mar 2019

Published

10 Sep 2020

Term ends

Not estimatedThe source records don’t support an estimate.

Classification20 codes

G10L 15/22
Speech recognition: procedures used during a speech recognition process, e.g. man-machine dialogue
G16H 10/20
ICT specially adapted for the handling or processing of patient-related medical or healthcare data: for electronic clinical trials or questionnaires
G10L 15/02
Speech recognition: feature extraction for speech recognition; Selection of recognition unit
G10L 15/08
Speech recognition: speech classification or search

Abstract

Claims18