# US 2026/0228420 A1 · Systems and methods for transcript keywords

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0228420 A1
- Document ID: US20260228420A1
- Status: Published application
- Kind: A1
- Application number: 19/465,988
- Filed: 2026-01-30
- Published: 2026-08-06
- Earliest claimed: 2025-02-03
- Applicants: Jpmorgan Chase Bank, N.A.
- Inventors: Devin MOORE; Udayalakshmi RAMAN
- Web page: https://patentgrep.com/patent/US20260228420A1

## CPC

- G06F 40/169: Annotation, e.g. comment data or footnotes
- G06F 40/103: Formatting, i.e. changing of presentation of documents (automatic justification G06F40/189; automatic line break hyphenation G06F40/191)
- G06F 40/166: Editing, e.g. inserting or deleting
- G06F 40/216: using statistical methods
- G06F 40/247: Thesauruses; Synonyms
- G06F 40/258: Heading extraction; Automatic titling; Numbering
- G06F 40/279: Recognition of textual entities
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/30: Semantic analysis
- G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30)
- G06F 40/56: Natural language generation
- G10L 15/26: Speech to text systems (G10L15/08 takes precedence)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

In some embodiments, the techniques described herein relate to a method including: receiving an audio or video file in real-time or from a database; executing a closed caption service application programming interface (“API”), generating a closed caption feed, conducting a word lookup for a word and/or phrase of the closed caption feed, generating a word link to a database for the word and/or phrase of the closed caption feed, seeding a regular expression with an insert link with a related word and/or a related phrase, and generating the annotated feed based on the insert link

## Claims

### Claim 1

A computer-implemented method for generating annotated transcript text, comprising:
receiving, by a closed caption service application executed by one or more processors, a audio content or a video content as a stream or batch file;
generating, by the closed caption service application, a closed caption feed from the audio content or the video content;
transmitting, by an application programming interface (“API”) service, portions of the closed caption feed to a machine learning engine;
determining, by the machine learning engine, a topic based on semantic similarity among located keywords in the portions of the closed caption feed;
conducting, by a word search module, a word lookup for a word or a phrase of the closed caption feed against a database;
generating, by an expression module, a word link to the database for the word or the phrase of the closed caption feed, the word link including an index identifier;
seeding, by the closed caption service application, a regular expression with an insert link associated with a related word or a related phrase determined using the topic; and
generating, by the closed caption service application, an annotated feed by applying the regular expression to the closed caption feed to insert an adornment linked by the insert link.

### Claim 2 (depends on 1)

The method of claim 1, wherein the receiving comprises caching the audio content or the video content in contiguous segments of a predetermined duration and updating the closed caption feed on a segment boundary to support real-time processing.

### Claim 3 (depends on 1)

The method of claim 1, wherein the determining of the topic comprises computing a semantic similarity score among at least four consecutive words grouped as a phrase and selecting the topic when the semantic similarity score exceeds a threshold.

### Claim 4 (depends on 1)

The method of claim 1, further comprising correcting, by the machine learning engine, a transcription error detected in the closed caption feed by semantically searching a candidate replacement for a single word and separately for a multi-word phrase, and updating the annotated feed with a corrected text associated with the transcription error prior to seeding the regular expression.

### Claim 5 (depends on 1)

The method of claim 1, wherein the generating of the word link comprises assigning, for the word or the phrase, a pointer to a record or portion of a record in the database, the record storing one or more of a definition, a synonym set, a topic identifier, a color code, and a hyperlink target.

### Claim 6 (depends on 1)

The method of claim 1, wherein the insert link comprises a node in a linked list that encodes an adornment type, a start index, an end index, and a priority value, and wherein generating the annotated feed comprises resolving conflicts among overlapping nodes by applying the priority value.

### Claim 7 (depends on 1)

The method of claim 1, wherein the adornment comprises one or more of a highlighting, an underline, an italic, a bold, a title formatting, a section header formatting, an insertion of an emoji, an insertion of an image, and an insertion of a hyperlink.

### Claim 8 (depends on 1)

The method of claim 1, further comprising selecting, by the closed caption service application, a color palette progression to visually distinguish consecutive topics and subtopics, and applying the color palette to the adornment such that adjacent topics receive different color codes sequentially according to the color palette progression.

### Claim 9 (depends on 1)

The method of claim 1, wherein the regular expression encodes a search pattern for matching a topic-specific phrase and replacing the matched phrase with a formatted string comprising an original text of the closed caption feed and the insert link, and wherein the formatted string is updated when the topic changes in subsequent segments of the closed caption feed.

### Claim 10 (depends on 1)

The method of claim 1, further comprising updating a previously inserted annotation by replacing an acronym with an expanded phrase and optionally appending a parenthetical acronym, or converting a spelled-out number to a numeric representation, based on a rule set retrieved via the word link.

### Claim 11 (depends on 1)

The method of claim 1, wherein the annotated feed is generated as a stream and comprises a sequence of frames, a frame of the sequence of frames including a timestamp, a text buffer, and a list of insert links applied to the text buffer, and wherein the closed caption service application revises insert links in prior frames when the machine learning engine updates the topic.

### Claim 12 (depends on 1)

The method of claim 1, wherein the word lookup accesses an external database hosted in a cloud environment and a local cache, and further comprising synchronizing, by the word search module, the local cache with the external database based on a cache miss event for the closed caption feed.

### Claim 13 (depends on 1)

The method of claim 1, further comprising detecting, by the machine learning engine, a subtopic within the topic using a hierarchical clustering of keyword embeddings, and seeding a second regular expression with an insert link specific to the subtopic, the insert link applying a distinct adornment from a topic-level insert link.

### Claim 14 (depends on 1)

The method of claim 1, wherein the closed caption service application applies a spacing formatting and a header formatting to transform a phrase designated as a title header or a section header into a structural element in the annotated feed, and indexes the structural element via the index identifier.

### Claim 15 (depends on 1)

The method of claim 1, further comprising computing, by the machine learning engine, a confidence score for an insert link based on signal quality of the audio content or the video content and transcription certainty, and conditionally applying the insert link to the annotated feed only when the confidence score exceeds a threshold, otherwise flagging the location for deferred review.
