Sliding Window Speech Separation for Real-Time Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current speech separation systems, particularly those using the DUET method, face inefficiencies due to the complexity of processing voice data, leading to delayed recognition in multi-speaker environments, such as in vehicles, where quick separation of speech signals is necessary to avoid erroneous operations.

Innovation Solution

A speech separation system and method that employs a sliding window to preprocess speech signals by determining specific amplitude thresholds, selecting segments of the signal for processing, and transmitting these segments to a DUET module for separation, thereby reducing the data load and processing time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the DUET algorithm processes the entire speech segment directly, then the speech separation can be performed, but the processing time becomes excessively long

Engineering Contradiction:
Improvespeech separation accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies segmentation by dividing the entire speech signal into multiple short segments using a sliding window approach. Each segment is processed independently by the DUET algorithm, which significantly reduces the processing time for each segment while maintaining overall separation accuracy. The segments are then reassembled to produce the final separated speech signals.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements preliminary action by pre-processing the speech signal to identify and select only the most relevant segments for processing. This involves analyzing the speech signal characteristics beforehand to determine which segments contain actual speech content and should be processed, while skipping silent or low-quality segments, thereby reducing unnecessary processing time.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If the DUET algorithm processes the entire 4-second speech segment, then complete speech separation is achieved, but the system responds too slowly for real-time applications

Engineering Contradiction:
Improvespeech separation completenessVSAvoidresponse speed
Core Design Contradiction:
ReliabilityVSSpeed

Solution Approach 1:

The patent segments the 4-second speech into shorter intervals (e.g., 0.5-2 seconds per segment) processed sequentially or in parallel. This segmentation enables the system to process speech in smaller, manageable chunks that can be handled within real-time constraints while maintaining complete separation through cumulative processing of all segments.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs dynamic processing by adjusting the sliding window size and processing parameters based on the detected speech activity and signal characteristics. The system dynamically adapts the segment duration, window shift, and processing intensity to match the actual speech patterns, enabling real-time response while maintaining separation completeness.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If the system waits for the entire speech segment before processing, then the DUET algorithm can process complete data, but the recognition operation is delayed

Engineering Contradiction:
Improvespeech signal processing accuracyVSAvoidrecognition delay
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by continuously monitoring and pre-processing speech segments as they arrive, rather than waiting for the complete 4-second segment to finish before initiating processing. The system prepares and processes segments in advance or in overlapping time windows, reducing the overall delay while maintaining processing accuracy through continuous refinement.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements continuous processing by maintaining an ongoing sliding window that continuously processes speech segments as they arrive, rather than batch processing complete segments. This continuous action allows the system to recognize and respond to speech commands as they are spoken, eliminating delays associated with waiting for complete segments while maintaining accurate separation through continuous signal analysis.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP3935632B1Method and system for speech separation
Publication Date: 2024.04.24 HARMAN INT IND INC
  • EP3935632B1 patent drawingFigure 1
  • EP3935632B1 patent drawingFigure 2~3
  • EP3935632B1 patent drawingFigure 4~5

AI summary

The present disclosure is directed to a speech separation method and system using a sliding window. The method comprises: acquiring at least one speech from at least one user by at least one microphone and storing the at least one speech as a speech signal in a sound recording module; extracting the speech signal from the sound recording module and processing the extracted speech signal through a sliding window; and transmitting the processed speech signal to a DUET module for speech separation.