Voice Recognition Synchronization Using Pre-Action Buffering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Automatic speech recognition systems face challenges in synchronizing with user voice sequences in 'hands-free' environments, leading to recognition errors due to imperfect synchronization between the user's actions and the start and end of vocal sequences.

Innovation Solution

The method involves processing voice segments from a given instant preceding the user's action, using a delay line to retroactively transmit segments and validate voice activity detections within specific time intervals, ensuring accurate recognition without introducing significant delays or increased computational load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the system uses Push-to-Talk device to trigger voice recognition processing, then false alerts are avoided and recognition accuracy improves, but voice sequences that start before or continue after the key action are truncated causing recognition errors

Engineering Contradiction:
Improverecognition accuracyVSAvoidvoice sequence truncation
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The system performs preliminary voice activity detection and buffers voice segments before the actual recognition processing is triggered. When the Push-to-Talk action occurs, the system already has pre-captured voice data ready, allowing it to process the complete voice sequence without truncation at the beginning. This resolves the contradiction by preparing the system in advance to capture the full voice input.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts the recognition processing timing based on detected voice activity rather than strictly following the Push-to-Talk trigger moment. Voice activity detection continuously monitors the input stream and identifies when actual speech occurs, allowing the system to start processing at the appropriate moment regardless of when the key was pressed. This dynamic adjustment prevents both premature and delayed processing that would cause truncation.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If the system processes voice segments from a given instant preceding the user's action, then recognition errors due to late triggering are reduced, but processing delay increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing delay
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs partial processing by buffering only a limited time window of voice segments preceding the trigger action, rather than processing from the very beginning of potential speech. This selective buffering of a reasonable time interval (not excessively long) captures the necessary pre-action voice data while limiting the processing delay to an acceptable level. The system applies excessive action in capturing data but controls the processing scope to balance accuracy and delay.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system uses voice activity detection to automatically identify when actual speech occurs within the buffered period, eliminating the need for manual timing adjustment. The voice activity detection mechanism self-regulates the processing start time based on actual speech patterns, allowing the system to utilize pre-captured segments effectively without introducing unnecessary delays. This self-service approach optimizes the balance between using historical data and maintaining real-time performance.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP1866909B1Method for synchronization between a voice recognition processing operation and an action triggering said processing
Publication Date: 2018.04.04 ORANGE SA
  • EP1866909B1 patent drawingFigure 1(a)~2
  • EP1866909B1 patent drawingFigure 3~4(b)

AI summary

The invention concerns a method for synchronization between a processing operation by automatic speech recognition of a voice sequence (Sv) emitted by a speaker and at least one action of said speaker designed to trigger said processing. The invention is characterized in that the processing operation is performed as from a given time (t0) preceding said speaker's action. The invention is applicable to automatic speech recognition.