Systems and methods for GPT guided neural punctuation for conversational speech

A computer system comprising:

• one or more processors;
• and
• one or more hardware storage devices that store instructions that are executable by the one or more processors to cause the computer system to:
• obtain audio data comprising language utterances from a speaker;
• continuously decode the audio data to generate decoded audio data;
• after determining a linguistic boundary exists within an initial segment of the decoded audio data, identify a first portion of the initial segment and a second portion of the initial segment based on the linguistic boundary;
• output the first portion of the initial segment of the decoded audio data, wherein the first portion is determined to be a first complete sentence;
• determine that the second portion is an incomplete sentence and that a subsequent segment of the decoded audio data will facilitate formation of a second complete sentence when the subsequent segment is appended to the second portion;
• reduce latency in forming the second complete sentence by storing the second portion of the initial segment in a cache;
• in response to determining that a combination of the second portion and the subsequent segment, if combined, would form the second complete sentence, append the subsequent segment to the second portion of the initial segment to form the second complete sentence;
• identify a disfluency in the decoded audio data;
• generate a particular correction to correct the disfluency;
• apply the particular correction to the decoded audio data;
• and
• generate an updated decoded audio data reflecting the particular correction, the updated decoded audio data having improved readability over the decoded audio data for the spoken language utterance.

One or more hardware storage devices that store instructions that are executable by one or more processors to cause the one or more processors to:

• obtain audio data comprising language utterances from a speaker;
• continuously decode the audio data to generate decoded audio data;
• after determining a linguistic boundary exists within an initial segment of the decoded audio data, identify a first portion of the initial segment and a second portion of the initial segment based on the linguistic boundary;
• output the first portion of the initial segment of the decoded audio data, wherein the first portion is determined to be a first complete sentence;
• determine that the second portion is an incomplete sentence and that a subsequent segment of the decoded audio data will facilitate formation of a second complete sentence when the subsequent segment is appended to the second portion;
• reduce latency in forming the second complete sentence by storing the second portion of the initial segment in a cache;
• in response to determining that a combination of the second portion and the subsequent segment, if combined, would form the second complete sentence, append the subsequent segment to the second portion of the initial segment to form the second complete sentence;
• identify a disfluency in the decoded audio data;
• generate a particular correction to correct the disfluency;
• apply the particular correction to the decoded audio data;
• and
• generate an updated decoded audio data reflecting the particular correction, the updated decoded audio data having improved readability over the decoded audio data for the spoken language utterance.