Voice Detection Using Multi-Domain Audio Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current voice detection methods suffer from low accuracy due to reliance on single characteristics, which are sensitive to noise, leading to interference between voice signals and noise, resulting in reduced detection precision and efficiency in human-computer interaction.

Innovation Solution

A voice detection method that divides audio signals into segments and extracts both time domain and frequency domain characteristics to accurately identify target voice segments, reducing noise interference and improving detection accuracy by using multiple characteristics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single characteristic is extracted from the input signal for voice detection, then the detection process is simple, but the detection accuracy reduces due to sensitivity to noise and inability to distinguish interference sounds

Engineering Contradiction:
Improvedetection process complexityVSAvoidvoice detection accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The audio signal is divided into multiple audio frames, and each frame is further segmented into multiple sub-frames. This segmentation allows the system to analyze different characteristics from different segments, improving detection accuracy by examining multiple aspects of the signal rather than relying on a single characteristic from the entire signal.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts characteristics from multiple dimensions including time domain (energy, zero-crossing rate), frequency domain (spectral centroid, spectral rolloff), and temporal features. By adding these additional dimensional analyses, the system overcomes the limitation of single-characteristic detection and achieves better noise discrimination.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple characteristics in different domains are extracted and used for detection, then voice detection accuracy increases by reducing noise interference, but the detection process complexity increases

Engineering Contradiction:
Improvevoice detection accuracyVSAvoiddetection process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines multiple characteristics from different domains (time domain energy, zero-crossing rate, frequency domain spectral features) into a unified detection framework. By merging these complementary features, the system achieves robust voice detection that leverages the strengths of each characteristic while compensating for their individual weaknesses.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system dynamically adjusts detection parameters and thresholds based on the analyzed characteristics. By changing parameters adaptively according to the extracted features, the system maintains high detection accuracy while managing computational complexity through intelligent parameter selection rather than exhaustive processing.

Inventive Principle:
Principle #35Parameter changes

3Ease of operation

If traditional voice detection methods are used, then the system response requires manual button pressing, but this reduces human-computer interaction efficiency

Engineering Contradiction:
Improveoperation convenienceVSAvoidhuman-computer interaction efficiency
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The voice detection system enables the computer to automatically detect and respond to user commands without requiring manual initiation. The system continuously monitors audio input and autonomously identifies voice segments, allowing the computer to serve itself by detecting user intent directly, thereby eliminating the need for manual button pressing and improving interaction efficiency.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3447769B1Voice detection method and apparatus, and storage medium
Publication Date: 2022.03.30 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3447769B1 patent drawingFigure 1~3
  • EP3447769B1 patent drawingFigure 4
  • EP3447769B1 patent drawingFigure 5

AI summary

Embodiments of the present invention provide a voice detection method and apparatus and a storage medium. The method includes: dividing a to-be-detected audio signal into a plurality of audio segments (S302); extracting an audio characteristic in each audio segment of the plurality of audio segments, where the audio characteristic includes at least a time domain characteristic and a frequency domain characteristic of the audio segment (S304); and detecting at least one target voice segment from the plurality of audio segments according to the audio characteristic of each audio segment (S306). The embodiments of the present invention solve a technical problem that voice detection accuracy is relatively low in a related voice detection method.