Voice Recognition Wake-Up Word Segmentation and Silence Labeling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current voice recognition systems face inefficiencies due to the need to process entire user voices, including irrelevant information, which reduces recognition speed and accuracy.

Innovation Solution

A method where the wake-up voice is identified and labeled with a silence identifier, allowing it to be ignored during recognition, thereby reducing data processing and improving efficiency using a decoding algorithm.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the server recognizes the entire user voice to ensure complete information processing, then the recognition accuracy is improved, but the data processing volume increases and recognition efficiency decreases

Engineering Contradiction:
Improverecognition accuracyVSAvoidrecognition efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the user voice into two distinct parts: the wake-up voice and the actual instruction voice. By identifying and separating the wake-up voice (which contains irrelevant information for the recognition task), the system can process only the relevant instruction portion, thereby reducing data processing volume while maintaining recognition accuracy for the actual command.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and removes the wake-up voice from the overall user voice input. By taking out the irrelevant wake-up voice segment and labeling it with a silence identifier, the system extracts only the necessary instruction information for processing, reducing the data amount that needs to be analyzed by the decoding algorithm while preserving the accuracy of recognizing the actual user instruction.

Inventive Principle:
Principle #2Taking out (Extraction)

2Loss of information

If the server processes the entire user voice including wake-up words, then no information is lost, but the data amount to be processed increases reducing efficiency

Engineering Contradiction:
Improveinformation completenessVSAvoidprocessing efficiency
Core Design Contradiction:
Loss of informationVSProductivity

Solution Approach 1:

The patent performs preliminary action by identifying and labeling the wake-up voice with a silence identifier before the main recognition process. This preliminary segmentation and labeling step allows the system to pre-determine which portions of the audio should be ignored during decoding, ensuring that no relevant information is lost while significantly reducing the data volume that requires full processing.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies local quality by treating different parts of the user voice differently. The wake-up voice portion is marked with a silence identifier indicating it should be ignored, while the instruction voice portion maintains its full processing quality. This localized differentiation allows the system to maintain information completeness for relevant segments while improving processing efficiency by excluding irrelevant segments.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS11398228B2Voice recognition method, device and server
Publication Date: 2022.07.26 BEIJING BAIDU NETCOM SCI & TECH CO LTD
  • US11398228B2 patent drawing
  • US11398228B2 patent drawing

AI summary

A voice recognition method, device, and a server are provided. The method includes: receiving a user voice; determining a wake-up voice of a wake-up word in the user voice, according to an acoustic feature of the user voice; and labeling the wake-up voice with a silence identifier; and ignoring the wake-up voice based on the silence identifier during voice recognition. As such, when a complex decoding algorithm is used to recognize the user voice, recognition of the wake-up word that is irrelevant to an instruction of the user is omitted, thus reducing the data amount to be processed by the decoding algorithm and improving the efficiency of voice recognition.