Speech End-Pointing Using Speaker Recognition and Energy Features

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional end-pointing algorithms for speech signals struggle to accurately detect the start and end points of user speech, especially when it overlaps with background babble noise, leading to inaccurate processing and potential errors in virtual assistants.

Innovation Solution

A process that combines two likelihoods to identify the start or end point of speech signals: a first likelihood based on energy features and a second likelihood determined through speaker authentication, using both frame and segment analysis to distinguish user speech from babble noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional end-pointing algorithms use energy features to detect speech, then the detection process is simple and fast, but the accuracy deteriorates when babble noise is present

Engineering Contradiction:
Improvedetection speedVSAvoidend-point detection accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent combines energy-based detection (first likelihood) with speaker authentication (second likelihood) into a unified end-pointing system. The energy-based approach provides fast initial detection while speaker authentication adds robustness against babble noise, achieving both speed and accuracy through feature fusion

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces speaker authentication as an intermediary mechanism between raw energy detection and final end-point determination. This intermediary layer verifies whether detected speech corresponds to an authorized user, filtering out false positives from babble noise while maintaining detection efficiency

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If speaker authentication is performed on every audio segment, then the accuracy of user speech detection is improved, but the computational complexity increases

Engineering Contradiction:
Improveuser speech detection accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the audio stream into frames and further segments for selective speaker authentication. By processing audio at multiple hierarchical levels (frame-level energy detection, segment-level speaker verification), the system reduces computational load compared to analyzing every sample uniformly

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies speaker authentication selectively rather than continuously - performing it on segments where energy-based detection indicates potential speech presence. This partial application of the computationally intensive authentication process maintains accuracy while reducing overall processing complexity

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3158561B1Robust end-pointing of speech signals using speaker recognition
Publication Date: 2020.04.15 APPLE INC
  • EP3158561B1 patent drawingFigure 1
  • EP3158561B1 patent drawingFigure 2A~2D
  • EP3158561B1 patent drawingFigure 3

AI summary

Systems and processes for robust end-pointing of speech signals using speaker recognition are provided. In one example process, a stream of audio having a spoken user request can be received. A first likelihood that the stream of audio includes user speech can be determined. A second likelihood that the stream of audio includes user speech spoken by an authorized user can be determined. A start-point or an end-point of the spoken user request can be determined based at least in part on the first likelihood and the second likelihood.