Granted Patent B1

US 12,229,313 · App. 18/778,183 · Granted Feb 18, 2025

Inventors: Tejas Shastry (Chicago, IL); Anthony Tassone (St. Charles, IL); Corey Burrows (Lisle, IL)

Assignee: Truleo, Inc.

G06F21/6245G06V20/41G10L17/02G10L17/04G10L25/57G11B27/031

Abstract

Claims (72)

1. An apparatus, comprising:

a processor; and

a memory operably coupled to the processor, the memory storing instructions to cause the processor to:

receive video data that includes audio data,

provide the audio data as input to a first machine learning model to produce transcription data,

provide the audio data as input to a second machine learning model to produce speaker separation data,

segment the transcription data based on the speaker separation data to produce speaker separated transcription data,

provide a first portion of the speaker separated transcription data from a plurality of portions of the speaker separated transcription data as input to a context window of a third machine learning model to produce an attribute indication, the first portion (1) being associated with a first speaker from a plurality of speakers and (2) not being associated with remaining speakers from the plurality of speakers,

in response to producing the attribute indication, cause display of video data at a user compute device, the video data being associated with the first portion of the speaker separated transcription data and not remaining portions of the speaker separated transcription data from the plurality of portions of the speaker separated transcription data,

provide at least one of the first portion of the speaker separated transcription data or a second portion of the speaker separated transcription data as input to a fourth machine learning model to identify personal identifiable information (PII) text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data, the second portion being associated with a second speaker from the plurality of speakers, and

in response to identifying the PII text:

replace the PII text with redaction text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data, and

cause display of the transcription data including the redaction text at the user compute device.

2. The apparatus of claim 1 , wherein:

the first speaker is a peace officer; and

the second speaker is a civilian.

3. The apparatus of claim 1 , further comprising an interface configured to communicate with a body camera, the video data being received from the body camera.

4. The apparatus of claim 1 , wherein:

the third machine learning model includes (1) a first large language model (LLM) configured to receive the first portion of the speaker separated transcription data as input to produce a speaker specific attribute indication and (2) a second LLM configured to receive the first portion of the speaker separated transcription data and the second portion of the speaker separated transcription data as input to produce a conversation attribute.

5. The apparatus of claim 1 , wherein the memory further stores instructions to cause the processor to:

receive a confirmation indication from the user compute device in response to the processor causing display of the video data; and

generate a label for the video data, the label being associated with the attribute indication.

6. The apparatus of claim 5 , wherein the memory further stores instructions to cause the processor to cause a training signal to be sent to retrain the third machine learning model based on the confirmation indication.

7. The apparatus of claim 1 , wherein the memory further stores instructions to cause the processor to:

receive a rejection indication from the user compute device in response to the processor causing display of the video data; and

cause a training signal to be sent to retrain the third machine learning model based on the rejection indication.

8. The apparatus of claim 1 , wherein the memory further stores instructions to cause the processor to:

cause display of the first portion of the speaker separated transcription data at the user compute device;

in response to producing the attribute indication, cause display of a tag that (1) is associated with the attribute indication and (2) highlights the first portion of the speaker separated transcription data at the user compute device;

cause display of an identifier associated with the first speaker; and

cause display of a selectable element associated with at least one of a confirmation indication or a rejection indication.

9. The apparatus of claim 1 , wherein the instructions to cause the processor to cause display of the video data at the user compute device include instructions to cause the processor to add the video data to a selectable queue that is associated with the attribute indication.

10. A method, comprising:

receiving, via a processor, audio data;

providing, via the processor, the audio data as input to a first machine learning model to produce transcription data;

providing, via the processor, the audio data as input to a second machine learning model to produce speaker separation data;

segmenting, via the processor, the transcription data based on the speaker separation data to produce speaker separated transcription data;

providing a first portion of the speaker separated transcription data from a plurality of portions of the speaker separated transcription data as input to a context window of a third machine learning model to produce an attribute indication, the first portion (1) being associated with a first speaker from a plurality of speakers and (2) not being associated with remaining speakers from the plurality of speakers;

in response to producing the attribute indication, causing display of video data at a user compute device, the video data being associated with the first portion of the speaker separated transcription data and not the remaining portions of the speaker separated transcription data;

providing, via the processor, at least the first portion of the speaker separated transcription data or a second portion of the speaker separated transcription data as input to a fourth machine learning model to identify personal identifiable information (PII) text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data, the second portion being associated with a second speaker from the plurality of speakers;

replacing, via the processor, the PII text with redaction text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data; and

causing, via the processor, display of the transcription data including the redaction text at the user compute device.

11. The method of claim 10 , wherein the redaction text is first redaction text, the method further comprising:

receiving, via the processor and from the user compute device, a redaction request associated with a portion of the transcription data; and

replacing, via the processor, the portion of the transcription data with second redaction text.

12. The method of claim 10 , wherein the causing display of the transcription data including the redaction text at the user compute device includes causing display of the transcription data such that the user compute device is unable to revert the redaction text to view the PII text via the user compute device.

13. The method of claim 10 , wherein the providing the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data as input to the fourth machine learning model includes:

generating, via the processor, a plurality of segments based on the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data; and

iteratively providing, via the processor, each segment from the plurality of segments as input to a context window of the fourth machine learning model to identify the PII text.

14. The method of claim 10 , wherein the audio data is included in video data that is generated by a body worn camera.

15. A method, comprising:

receiving, via a processor, audio data;

providing, via the processor, the audio data as input to a first machine learning model to produce transcription data;

providing, via the processor, the audio data as input to a second machine learning model to produce speaker separation data;

segmenting, via the processor, the transcription data based on the speaker separation data to produce speaker separated transcription data;

providing, via the processor, a first portion of the speaker separated transcription data from a plurality of portions of the speaker separated transcription data as input to a third machine learning model to produce an attribute indication, the first portion of the speaker separated transcription data (1) being attributable to a law enforcement officer from a plurality of speakers and (2) not being associated with remaining speakers from the plurality of speakers;

causing, via the processor, display of video data at a user compute device, the video data being associated with the first portion of the speaker separated transcription data and not remaining portions of the speaker separated transcription data from the plurality of portions of the speaker separated transcription data;

providing, via the processor, at least the first portion of the speaker separated transcription data or a second portion of the speaker separated transcription data as input to a fourth machine learning model to identify personal identifiable information (PII) text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data, the second portion being attributable to a community member;

replacing, via the processor, the PII text with redaction text in the at least one of the first portion of the speaker separated transcription data or the second portion of the speaker separated transcription data; and

causing, via the processor, display of the transcription data including the redaction text at the user compute device.

16. The method of claim 15 , wherein the attribute indication is associated with a speaker specific attribute.

17. The method of claim 15 , further comprising:

receiving, via the processor and from the user compute device, an invalidation request;

deleting, via the processor, the attribute indication in response to the receiving the invalidation request; and

generating, via the processor, audit data that includes an indication of the invalidation request.

18. The method of claim 15 , further comprising:

receiving, via the processor, a confirmation indication from the user compute device in response to the causing display of the video data; and

associating, via the processor, a label with the video data, the label being associated with the attribute indication.

19. The method of claim 15 , further comprising causing, via the processor, a training signal to be sent to retrain the third machine learning model based on a user confirmation of the attribute indication.

20. The method of claim 15 , wherein:

the providing the first portion of the speaker separated transcription data as input to the third machine learning model includes providing the second portion of the speaker separated transcription data as input to the third machine learning model to produce the attribute indication, the attribute indication being a conversation attribute indication.

21. The method of claim 20 , wherein the third machine learning model includes (1) a first large language model (LLM) configured to receive the first portion of the speaker separated transcription data as input to produce a speaker specific attribute indication and (2) a second LLM configured to receive the second portion of the speaker separated transcription data as input to produce the conversation attribute indication.

Assignments (1)

Continuity (1)

References Cited (153)

Cited By (5)

US 12,489,778 US 12,499,243 US 12,537,750 US 12,609,968 US 12,694,106