Dynamic Language Model Adaptation for Voice Search Queries

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional voice search systems lack personalization and relevance in converting voice inputs to text search queries, as they do not consider specific circumstances such as location, time, or user profiles, leading to less pertinent search results.

Innovation Solution

A voice search system that applies a customized language model by adjusting probability values based on geographic location, time, weather, and user profiles, using dynamic language modeling techniques to generate more relevant text search queries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a base language model is used for speech recognition, then the system can process voice inputs, but the search queries lack personalization and relevance to user context

Engineering Contradiction:
ImprovepersonalizationVSAvoidcontext information
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The system pre-processes user profiles, location data, and contextual information before speech recognition occurs. Language models are pre-customized based on user characteristics and historical behavior, so that when voice input is received, the personalized model is already ready to provide context-aware recognition without adding latency to the actual speech-to-text conversion process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The language model dynamically adapts its parameters based on real-time context factors such as user location, time of day, and current activity state. The system adjusts probability distributions and word sequences in the language model according to contextual conditions, enabling the same voice input to be interpreted differently based on situational factors without requiring manual reconfiguration.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If context-based customization is applied to language models, then search query relevance improves, but system complexity increases

Engineering Contradiction:
Improvequery accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The customization system is divided into separate modular components: user profile analysis module, context detection module, language model selection module, and probability adjustment module. Each module handles a specific aspect of customization independently, allowing the system to achieve high query accuracy through coordinated modular operations rather than a monolithic complex system.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of creating entirely separate language models for different contexts, the system adjusts parameters within existing language models based on contextual factors. By modifying probability values, word sequence weights, and recognition thresholds dynamically, the system achieves context-aware accuracy without the complexity of maintaining multiple complete model instances.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3754651B1Dynamic language model adaptation for generating voice-based search queries
Publication Date: 2024.05.22 GOOGLE LLC
  • EP3754651B1 patent drawingFigure 1
  • EP3754651B1 patent drawingFigure 2
  • EP3754651B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programes encoded on computer storage media for speech recognition. One of the methods comprises receiving a voice search query from a user device, the voice search query being associated with a query context; determining, based on the query context, that a customized language model is to be used to convert the voice search query to a text search query; customizing a base language model using one or more language model rules to generate a customized language model, the customizing comprising dynamically adjusting one or more probabilities in the base language model according to the one or more language model rules; using the customized language model to generate a text search query from the voice search query; and providing for display on the user device, one or more search results responsive to the text search query.