Providing intelligent transcriptions of sound messages in a messaging application
AssigneeFacebook, Inc.
InventorsGuillaume Lample, Arnaud Vincent Albalat
Life of this patent
Estimating the term…
Abstract
One or more embodiments described herein include methods and systems of creating transcribed electronic communications based on sound inputs. More specifically, systems and methods described herein provide users the ability to easily and effectively send an electronic communication that includes a textual message transcribed from a sound input. Additionally, systems and methods described herein provide an analysis of a textual message transcribed from a sound input allowing users to correct an inaccurate or incorrect transcription.
Claims20
A method for providing a textual transcription of a sound recording comprising:
- transcribing, by at least one processor, a recorded message into a textual message having a plurality of language elements;
- identifying a first language element from among the plurality of language elements of the textual message;
- determining, by the at least one processor, input values based on one or more features associated with the first language element and based on one or more features associated with additional language elements of the textual message;
- identifying a-posteriori confidence levels for the first language element and each additional language element of the additional language elements;
- providing the input values and the a-posteriori confidence levels to an input layer of a neural network;
- receiving, as an output of the neural network, an indication of whether the first language element of the transcribed textual message is correct;
- and
- based on the indication of whether the first language element of the transcribed textual message is correct, associating a confidence indicator with the first language element of the transcribed textual message.
A method for providing a textual transcription of a sound recording comprising:
- transcribing, by at least one processor, a recorded message into a textual message comprising a sentence with a plurality of words;
- identifying an analysis word from among the plurality of words of the sentence;
- identifying one or more contextual words associated with the analysis word;
- identifying a-posteriori confidence levels for the analysis word and the one or more contextual words associated with the analysis word;
- determining, by the at least one processor, input values for a first set of features associated with the analysis word, and a second set of features associated with the one or more contextual words;
- providing the input values and the a-posteriori confidence levels to an input layer of a neural network, wherein the input layer has a plurality of neurons based on a total number of features in the first set of features and the second set of features;
- receiving an output value from an output layer of the neural network, wherein the output value indicates whether the analysis word is correct;
- and
- based on the output value, associating a confidence indicator with the analysis word in the textual message.
A system for providing a textual transcription of a sound recording comprising:
- at least one processor;
- at least one non-transitory computer readable storage medium storing instructions thereon that, when executed by the at least one processor, cause the system to:
- transcribe a recorded message into a textual message comprising a sentence with a plurality of words;
- identify an analysis word from among the plurality of words of the sentence;
- identify one or more contextual words associated with the analysis word;
- identify a-posteriori confidence levels for the analysis word and the one or more contextual words associated with the analysis word;
- determine input values for a first set of features associated with the analysis word and a second set of features associated with the one or more contextual words;
- provide the input values and the a-posteriori confidence levels to an input layer of a neural network, wherein the input layer has a plurality of neurons based on a total number of features in the first set of features and the second set of features;
- receive an output value from an output layer of the neural network, wherein the output value indicates whether the analysis word is correct;
- and
- based on the output value, associate a confidence indicator with the analysis word in the textual message.
Dependencies are inferred from the claim text.