Context-Biased Speech Recognition for Atypical Utterances

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Speakers with atypical speech, such as apraxia, cluttering, or stuttering, face challenges in ensuring that others understand them due to decreased accuracy in speech recognition systems.

Innovation Solution

A system employs multiple speech recognizers, selectively activating an alternative speech recognizer based on detecting atypical speech, and biases it using context from a general recognizer's transcription to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a general speech recognizer is used to transcribe speech, then the system is simple and fast, but the transcription accuracy decreases when the speaker has atypical speech

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

Solution Approach 1:

The system dynamically switches between a general speech recognizer and an alternative speech recognizer based on detecting atypical speech characteristics. The alternative recognizer is activated only when atypical speech is detected, allowing the system to adapt its complexity based on the speech type while maintaining high accuracy for atypical speakers

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the operational parameters by selecting different recognizer models based on speech characteristics. When atypical speech is detected, the system transitions to using an alternative speech recognizer with different processing parameters optimized for atypical speech patterns, thereby improving accuracy without permanently increasing system complexity

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple speech recognizers are employed to handle atypical speech, then the transcription accuracy improves, but the system complexity increases

Engineering Contradiction:
Improvetranscription accuracyVSAvoidnumber of speech recognizers
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The speech recognition system is segmented into two distinct components: a general speech recognizer for typical speech and an alternative speech recognizer for atypical speech. This segmentation allows each component to be optimized for its specific function, improving overall accuracy while managing complexity through functional separation

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A speech classifier acts as an intermediary between the audio input and the speech recognizers. The classifier analyzes speech characteristics and directs the audio stream to the appropriate recognizer, enabling the system to handle multiple speech types accurately without requiring all recognizers to process all inputs, thus managing complexity

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If the alternative speech recognizer is always activated, then the accuracy for atypical speech is maintained, but the processing speed and efficiency decrease

Engineering Contradiction:
Improvetranscription accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system dynamically activates the alternative speech recognizer only when atypical speech is detected by the speech classifier. For typical speech, the faster general speech recognizer is used, maintaining high processing efficiency. When atypical speech is detected, the system switches to the alternative recognizer to ensure accuracy, thus optimizing both speed and accuracy based on real-time speech characteristics

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4085452B1Speech recognition
Publication Date: 2025.12.31 GOOGLE LLC
  • EP4085452B1 patent drawingFigure 1
  • EP4085452B1 patent drawingFigure 2
  • EP4085452B1 patent drawingFigure 3

AI summary

A method (400) includes receiving acoustic features (205) of a first utterance (108) spoken by a first user (102) that speaks with typical speech and processing the acoustic features of the first utterance using a general speech recognizer (124) to generate a first transcription (218) of the first utterance. The operations also include analyzing the first transcription of the first utterance to identify one or more bias terms (224) in the first transcription and biasing an alternative speech recognizer (132) on the one or more bias terms. The operations also include receiving acoustic features of a second utterance spoken by a second user that speaks with atypical speech and processing, using the alternative speech recognizer biased on the one or more terms identified in the first transcription, the acoustic features of the second utterance to generate a second transcription (230) of the second utterance.