Dynamic Speech Recognition Thresholds During Audio Rendering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech processing systems in automated assistants struggle with accurately recognizing spoken commands due to interference from audio output and background noise, leading to misrecognition and inefficient resource usage.

Innovation Solution

Dynamic speech processing adaptation is applied based on characteristics of the content being rendered, such as loudness and historical usage data, to bias the recognition of specific terms during relevant portions of audio playback, enhancing accuracy and robustness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If audio output is rendered by the device while receiving speech, then the automated assistant can provide audio feedback and entertainment content, but the speech processing accuracy deteriorates due to audio interference and background noise

Engineering Contradiction:

Improveaudio output capabilityVSAvoidspeech recognition accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts speech processing parameters based on the current audio output state. The speech processing engine modifies its recognition thresholds and sensitivity levels in response to detected audio characteristics, allowing it to adapt between high-accuracy mode (low audio output) and tolerant mode (high audio output), thus resolving the contradiction between audio versatility and speech accuracy

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes processing parameters such as speech threshold levels, attention weights, and recognition sensitivity based on the audio output characteristics. When audio output is detected, the system adjusts parameters to reduce false positives while maintaining the ability to recognize clear commands, thereby maintaining both audio capability and speech accuracy

Inventive Principle:

Principle #35Parameter changes

2Productivity

If speech processing continuously monitors audio data, then the automated assistant can respond to commands in real-time, but computational resources are wasted processing irrelevant audio segments

Engineering Contradiction:

Improvereal-time response capabilityVSAvoidcomputational resource consumption

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The system segments the audio stream into relevant segments based on detected audio characteristics and content type. Instead of processing all audio continuously, the speech processing engine identifies and processes only segments that contain potential commands, filtering out irrelevant segments such as music or background noise, thus maintaining real-time responsiveness while reducing computational waste

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary analysis of audio characteristics before full speech processing. By pre-filtering audio segments based on loudness, content type, and contextual relevance, the system prepares only the most promising segments for detailed speech recognition, reducing overall computational load while maintaining real-time response capability

Inventive Principle:

Principle #10Preliminary action

3Device complexity

If the speech processing engine uses fixed recognition thresholds, then the system is simple to implement, but it cannot adapt to varying audio conditions such as loudness changes and background noise

Engineering Contradiction:

Improvespeech processing simplicityVSAvoidspeech recognition robustness

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The system transitions from static to dynamic threshold settings that automatically adjust based on detected audio conditions. The speech processing engine continuously monitors audio characteristics such as overall loudness, background noise levels, and content type, then adapts recognition thresholds accordingly, maintaining robustness across varying conditions without requiring complex manual configuration

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system implements feedback mechanisms where the speech processing engine uses detected audio characteristics to adjust its recognition parameters in real-time. By monitoring the audio environment and feeding this information back into the processing system, the engine automatically optimizes its thresholds for current conditions, improving reliability while keeping the implementation relatively simple

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250285620A1Biasing speech processing based on audibly rendered content, including dynamically adapting over duration of rendering

Publication Date: 2025.09.11 GOOGLE LLC

AI summary

Implementations set forth herein relate to an automated assistant that can bias speech processing towards certain requests according to whether those requests are relevant to content that is being rendered, or is expected to be rendered, at a computing device. In this way, speech processing can be dynamically biased according to features of content that may be rendered by a particular application and/or a particular device. Biasing can be performed during rendering of a portion of content determined to be relevant to a particular request by adjusting a score threshold that is used for determining whether a particular request was received. When the portion of content is no longer being rendered, the threshold can return to a particular value, or be adjusted again according to a subsequent portion of the content.