Audio Caption Correction via False Alarm and Miss Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current audio captioning systems lack the ability to identify and correct false positives (words that should not be in the caption) and false negatives (missing words that should be in the caption).

Innovation Solution

The system generates a corrected caption by using a captioning model to create a candidate caption from input data, a tagging model to generate candidate tags, and a false alarm and miss detector to identify false alarms and false negatives. A caption correction engine then uses this information to produce a corrected caption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a captioning model generates candidate captions from input audio data, then the system can produce audio captions, but the captions contain false positives (words that should not be in the caption) and false negatives (missing words that should be in the caption)

Engineering Contradiction:
Improvecaption accuracyVSAvoiddetection of false alarms and misses
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent introduces an intermediary detection system that acts as a mediator between the captioning model and the final caption output. This intermediary component analyzes the relationship between audio tags and caption detections to identify false positives and false negatives, thereby improving caption accuracy without directly modifying the captioning model itself.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements a feedback mechanism where the detection results (false alarms and misses) are used to generate a corrected caption. This feedback loop allows the system to continuously improve caption quality by learning from its own errors, comparing detected elements against ground truth audio tags, and adjusting subsequent caption generation accordingly.

Inventive Principle:
Principle #23Feedback

2Reliability

If the system compares audio tags and detections to identify false alarms and misses, then caption accuracy improves, but the system complexity increases

Engineering Contradiction:
Improvecaption accuracyVSAvoidsystem architecture
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the caption evaluation process into distinct functional components: generating audio tags from audio data, generating candidate captions from audio data, detecting false positives by comparing tags and detections, and detecting false negatives by comparing tags and detections. This segmentation allows each component to be optimized independently and makes the overall system more manageable despite its increased complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs self-evaluation and self-correction by automatically comparing its own caption output against independently generated audio tags. This self-service capability allows the system to identify and correct its own errors without external intervention, justifying the added complexity through autonomous improvement.

Inventive Principle:
Principle #25Self-service

3Reliability

If the system generates corrected captions by identifying and correcting false alarms and misses, then the reliability of captioned audio data improves, but the processing time increases

Engineering Contradiction:
Improvecaption reliabilityVSAvoidcaption generation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary action by generating audio tags independently before caption generation and using these tags as a reference for detection. By preparing this ground truth data in advance, the system can efficiently perform false alarm and miss detection without adding significant processing time during the main caption generation workflow.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20250078828A1Automated audio caption correction using false alarm and miss detection
Publication Date: 2025.03.06 QUALCOMM INC
  • US20250078828A1 patent drawing
  • US20250078828A1 patent drawing
  • US20250078828A1 patent drawing

AI summary

Systems and techniques are provided for natural language processing. A system generates a plurality of tokens (e.g., words or portions thereof) based on input content (e.g., text and/or speech). The system searches through the plurality of tokens to generate a first ranking the plurality of tokens based on probability. The system generates natural language inference (NLI) scores for the plurality of tokens to generate a second ranking of the plurality of tokens based on faithfulness to the input content (e.g., whether the tokens produce statements that are true based on the input content). The system generates output text that includes at least one token selected from the plurality of tokens based on the first ranking and the second ranking.