Asynchronous Speech Model Retrieval and Pre-Caching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Distributed speech recognition systems face limitations due to the large size of additional speech recognition models, which consume storage space and increase latency when retrieved from storage servers, affecting user experience and accuracy.

Innovation Solution

Implementing asynchronous retrieval and caching of additional speech recognition models, allowing for parallel processing with base models and pre-caching based on user patterns to minimize latency and improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If additional speech recognition models are stored and retrieved from storage servers, then speech recognition accuracy is improved, but retrieval latency increases and storage space is consumed

Engineering Contradiction:
Improvespeech recognition accuracyVSAvoidretrieval latency
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by predicting which speech recognition models will be needed based on user patterns and usage history, then pre-retrieves and caches these models before they are actually requested. This proactive caching strategy ensures models are ready when needed, eliminating retrieval latency during actual speech recognition operations while maintaining high accuracy through access to specialized models

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If additional speech recognition models are stored and retrieved from storage servers, then speech recognition accuracy is improved, but storage space is consumed

Engineering Contradiction:
Improvespeech recognition accuracyVSAvoidstorage space
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system applies local quality by caching speech recognition models at specific locations (edge servers or client devices) based on local user patterns and usage characteristics. Instead of uniformly storing all models everywhere, the system selectively caches models locally where they are most needed, optimizing the balance between storage space consumption and speech recognition accuracy for each local context

Inventive Principle:
Principle #3Local quality

3Measurement precision

If additional speech recognition models are used, then speech recognition accuracy is improved, but system complexity increases

Engineering Contradiction:
Improvespeech recognition accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system implements self-service through automated model selection and caching mechanisms that use machine learning to predict user needs and automatically retrieve appropriate models without manual intervention. This automation reduces the operational complexity of managing multiple speech recognition models, as the system autonomously handles model selection, caching, and retrieval based on observed usage patterns

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP2932501B1Speech model retrieval in distributed speech recognition systems
Publication Date: 2023.05.17 AMAZON TECH INC
  • EP2932501B1 patent drawingFigure 1
  • EP2932501B1 patent drawingFigure 2
  • EP2932501B1 patent drawingFigure 3

AI summary

Features are disclosed for managing the use of speech recognition models and data in automated speech recognition systems. Models and data may be retrieved asynchronously and used as they are received or after an utterance is initially processed with more general or different models. Once received, the models and statistics can be cached. Statistics needed to update models and data may also be retrieved asynchronously so that it may be used to update the models and data as it becomes available. The updated models and data may be immediately used to re-process an utterance, or saved for use in processing subsequently received utterances. User interactions with the automated speech recognition system may be tracked in order to predict when a user is likely to utilize the system. Models and data may be pre-cached based on such predictions.