Dynamic Language Model for Voice Query Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Automated speech recognition systems often misrecognize subsequent voice queries due to static language models that fail to account for the inherent relationship between initial and subsequent queries submitted by the same user within a particular context, leading to misrecognition of low-quality queries with multiple candidate transcriptions.

Innovation Solution

The system identifies predicted follow-up queries associated with an initial query and adjusts the language model to increase the probabilities of these queries, ensuring that subsequent voice queries are more likely to be transcribed accurately by reflecting the user's context and query history.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a static language model is used for speech recognition, then the system structure is simple and easy to implement, but the voice recognition accuracy deteriorates for subsequent queries that should be predicted based on user context

Engineering Contradiction:
Improvevoice recognition accuracyVSAvoidlanguage model complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary analysis of user context and query patterns before processing subsequent voice queries. By pre-computing predicted follow-up queries based on the initial query and user history, the system prepares the adjusted language model in advance, improving recognition accuracy for anticipated query types without adding significant processing complexity during actual query handling.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The language model is transformed from a static structure to a dynamic one that adapts based on user context. The system adjusts language model probabilities dynamically by identifying predicted follow-up queries and modifying the language model to increase probabilities for context-relevant terms, allowing the model to flexibly respond to different user scenarios while maintaining a relatively simple base structure.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If the language model probabilities are adjusted dynamically based on user context, then the transcription accuracy for subsequent queries improves, but the processing complexity and computational resources increase

Engineering Contradiction:
Improvetranscription accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Instead of adjusting the entire language model globally, the system applies localized probability adjustments only to specific terms and query patterns relevant to the predicted follow-up queries. This targeted approach increases transcription accuracy for context-relevant terms while minimizing the computational overhead associated with processing the entire language model.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system changes specific probability parameters within the language model based on user context analysis. By modifying only the probability values for predicted follow-up query terms rather than restructuring the entire model, the system achieves improved transcription accuracy with relatively low computational cost, as parameter adjustment is more efficient than complete model reconstruction.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If predicted follow-up queries are identified and language model probabilities are increased for these queries, then subsequent queries are more likely to be transcribed correctly, but the system may overlook unexpected or novel user queries

Engineering Contradiction:
Improvetranscription reliabilityVSAvoidquery flexibility
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system applies partial adjustment to the language model by increasing probabilities for predicted follow-up queries while maintaining baseline probabilities for other query types. This partial action approach ensures that context-predicted queries receive enhanced transcription reliability, while unexpected or novel queries still have a chance to be recognized through the preserved baseline probability distribution, balancing reliability and versatility.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3469489B1Follow-up voice query prediction
Publication Date: 2022.11.16 GOOGLE LLC
  • EP3469489B1 patent drawingFigure 1
  • EP3469489B1 patent drawingFigure 2
  • EP3469489B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for predicting follow-up queries to an initial transcription of an utterance. In some implementations, one or more follow-up queries that are pre-associated with a transcription of an initial utterance of a user are identified. A new or modified language model in which a respective probability associated with one or more of the follow-up queries is increased with respect to an initial language model is obtained. Subsequent audio data corresponding to a subsequent utterance of the user is then received. The subsequent audio data is processed using the new or modified language model to generate a transcription of the subsequent utterance. The transcription of the subsequent utterance is then provided for output to the user.