Cross-Session Speech Hypothesis Retention for Misrecognition Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automated assistants often misrecognize spoken utterances, leading to prolonged human-to-computer dialogs and unnecessary computational resource consumption due to manual corrections by users.

Innovation Solution

Preserve multiple speech hypotheses across computing devices and dialog sessions, allowing automatic correction or user interaction to rectify misrecognitions, reducing the need for manual input and conserving resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If speech recognition is performed to generate a transcription, then the user can communicate with the automated assistant, but misrecognitions occur causing prolonged dialogs and additional computational resource consumption

Engineering Contradiction:

Improvedialog efficiencyVSAvoiddialog length

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent performs speech recognition in advance to generate multiple transcription hypotheses before the user interacts with the automated assistant. These hypotheses are stored and can be retrieved during the dialog, allowing the system to quickly present alternative interpretations without requiring the user to repeat themselves, thus reducing dialog length and improving efficiency

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system uses the generated transcription hypotheses as feedback to improve speech recognition accuracy. When misrecognitions are detected, the system can reference alternative hypotheses to correct errors, creating a feedback loop that continuously improves recognition accuracy and reduces the need for manual corrections, thereby shortening dialog length

Inventive Principle:

Principle #23Feedback

2Productivity

If speech recognition is performed to generate a transcription, then the user can communicate with the automated assistant, but misrecognitions occur causing additional computational resource consumption

Engineering Contradiction:

Improvedialog efficiencyVSAvoidcomputational resource consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

Instead of performing full speech recognition analysis every time a user speaks, the system uses pre-generated transcription hypotheses for common phrases or patterns. This partial recognition approach reduces computational resource consumption while maintaining dialog efficiency, as the system only performs full analysis when necessary rather than excessively processing every utterance

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system performs speech recognition in advance to generate multiple transcription hypotheses before the user interacts with the automated assistant. These hypotheses are stored and can be retrieved during the dialog, allowing the system to quickly present alternative interpretations without requiring the user to repeat themselves, thus reducing dialog length and improving efficiency

Inventive Principle:

Principle #10Preliminary action

Solution Approach 3:

Inventive Principle:

Principle #23Feedback

3Reliability

If multiple speech hypotheses are generated and stored, then misrecognitions can be corrected, but memory resources are consumed

Engineering Contradiction:

Improvespeech recognition accuracyVSAvoidmemory resources

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The patent extracts only the most relevant transcription hypotheses from the full set of possible interpretations and stores them for later use. Rather than storing all possible speech recognition outcomes, the system selectively retains hypotheses that are most likely to be useful for correction, reducing memory resource consumption while maintaining the ability to correct misrecognitions and improve reliability

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The system changes the parameter of hypothesis storage by storing hypotheses in a compressed or optimized format, or by storing them selectively based on confidence thresholds. This allows the system to maintain multiple speech hypotheses for correction purposes while managing memory resources efficiently, balancing reliability improvement with resource consumption

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentEP4128220B1Preserving speech hypotheses across computing devices and/or dialog sessions

Publication Date: 2026.04.01 GOOGLE LLC

Figure 1
Figure 2
Figure 3A

AI summary

Implementations can receive, at a computing device, audio data corresponding to a spoken utterance of a user, process the audio data to generate, for one or more parts of the spoken utterance, a plurality of speech hypotheses, select a given one of the speech hypotheses, cause the given one of the speech hypotheses to be incorporated as a portion of a transcription associated with the software application, and store the plurality of speech hypotheses. In some implementations, the plurality of speech hypotheses can be loaded at an additional computing device when the transcription is accessed at the additional computing device. In additional or alternative implementations, the plurality of speech hypotheses can be loaded into memory of the computing device when the software application is reactivated and/or when a subsequent dialog session associated with the transcription is initiated.