Neural Network Voice Activity Detection in Noisy Audio

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current speech recognition systems face challenges in accurately detecting voice activity in audio inputs, particularly in noisy conditions, which affects the performance of speech recognition systems.

Innovation Solution

A neural network-based voice activity detection system is trained using labeled audio waveforms to classify input audio as containing speech or not, employing a combination of convolutional, long short-term memory, and fully connected deep neural network layers to process audio waveforms and generate accurate classifications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional voice activity detection methods are used, then the system is simpler to implement, but the detection accuracy deteriorates in noisy conditions

Engineering Contradiction:
Improvevoice activity detection accuracyVSAvoidneural network system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The audio waveform is divided into multiple frames, with each frame processed independently through the neural network to generate voice activity predictions. This segmentation allows the system to handle noisy conditions frame-by-frame while maintaining overall detection accuracy through temporal aggregation of results.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system transforms the one-dimensional audio waveform into a two-dimensional representation by creating a spectrogram or feature matrix across time and frequency dimensions. This dimensional transformation enables the neural network to capture both temporal and spectral characteristics of speech, significantly improving detection accuracy in noisy environments.

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

2Reliability

If voice activity detection is performed continuously, then speech recognition performance is improved, but computational energy consumption increases

Engineering Contradiction:
Improvespeech recognition performanceVSAvoidcomputational energy consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system extracts only the relevant portions of audio containing voice activity using the trained neural network, rather than processing the entire audio stream continuously. By identifying and extracting speech segments based on voice activity detection, the system reduces computational energy consumption while maintaining speech recognition performance.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The neural network performs voice activity detection at selected intervals or on sampled frames rather than analyzing every single audio sample. This partial action approach provides sufficient detection accuracy to trigger speech recognition only when needed, reducing overall computational energy consumption while maintaining reliability.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3347896B1Voice activity detection
Publication Date: 2019.09.04 GOOGLE LLC
  • EP3347896B1 patent drawingFigure 1
  • EP3347896B1 patent drawingFigure 2
  • EP3347896B1 patent drawingFigure 3

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting voice activity. In one aspect, a method include actions of receiving, by a neural network included in an automated voice activity detection system, a raw audio waveform, processing, by the neural network, the raw audio waveform to determine whether the audio waveform includes speech, and provide, by the neural network, a classification of the raw audio waveform indicating whether the raw audio waveform includes speech.