Voice Assistant Active Listening With Non-Visual Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing assistant systems face challenges in increasing operating capacity while maintaining active listening capabilities, especially when the client system is locked, and providing non-interruptive feedback to users to avoid confusion about issuing voice commands.

Innovation Solution

The solution involves providing continuous non-visual feedback through subtle and non-interruptive background sounds, allowing the assistant system to access increased computing capacity without unlocking the client system, and using a hybrid architecture that leverages both client-side and server-side processes for efficient resource utilization and privacy protection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional NLP-based approaches are used for sentiment analysis, then text processing capability is provided, but contextual understanding and emotional nuance detection are insufficient

Engineering Contradiction:
Improvecontextual understanding accuracyVSAvoidsystem architecture complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments audio processing and text processing into separate specialized modules, with audio data processed by audio encoders and text data processed by text encoders, then combines their outputs. This segmentation allows each module to be optimized for its specific modality while maintaining overall system functionality.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces audio embeddings as an intermediary representation that captures emotional context from audio signals. These embeddings serve as a bridge between the audio input and the text analysis components, enabling the system to incorporate non-verbal emotional cues into the sentiment analysis process.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If audio and text data are processed separately, then processing efficiency is maintained, but emotional context integration is lost

Engineering Contradiction:
Improveemotional context retentionVSAvoidprocessing efficiency
Core Design Contradiction:
Loss of informationVSProductivity

Solution Approach 1:

The system merges audio embeddings and text embeddings into a unified combined embedding representation. This combination occurs at the feature level, allowing the model to leverage both audio-based emotional cues and text-based semantic information simultaneously for more accurate sentiment and emotion detection.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The dual-encoder architecture is designed to handle multiple types of input data (audio and text) through a unified processing framework. The same transformer-based architecture processes both modalities, demonstrating multi-functionality while maintaining processing efficiency through parallel operation of the separate encoders.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If comprehensive emotional analysis is implemented, then detection accuracy is improved, but computational resources and processing time increase

Engineering Contradiction:
Improveemotion detection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary processing by extracting audio embeddings and text embeddings separately through specialized encoders before combining them. This preliminary separation allows for optimized processing of each modality independently, reducing overall computation time while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent utilizes parameter changes in the embedding space to represent different emotional states. By transforming raw audio and text inputs into standardized embedding representations with specific dimensionalities and normalizations, the system enables efficient comparison and analysis across different input types while maintaining detection accuracy.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4327199B1Active listening for assistant systems
Publication Date: 2026.05.06 META PLATFORMS INC
  • EP4327199B1 patent drawingFigure 1
  • EP4327199B1 patent drawingFigure 2
  • EP4327199B1 patent drawingFigure 3

AI summary

In one embodiment, a method includes receiving a first user input comprising a wake word associated with an assistant xbot from a first client system, setting the assistant xbot into a listening mode, wherein a continuous non-visual feedback is provided via the first client system while the assistant xbot is in the listening mode, receiving a second user input comprising a user utterance from the first client system while the assistant xbot is in the listening mode, determining the second user input has ended based on a completion of the user utterance, and setting the assistant xbot into an inactive mode, wherein the non-visual feedback is discontinued via the first client system while the assistant xbot is in the inactive mode.