Voice Interaction Intent Recognition via Acoustic Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional voice interaction solutions are cumbersome, requiring repeated wake-up words for interaction and failing to accurately recognize user intentions, leading to poor user experience and incorrect interactions due to reliance on specific wake-up words and simple voice features.
Innovation Solution
A method and apparatus that extract acoustic features like filter bank or Mel frequency cepstrum coefficients from voice data to determine interaction intentions, eliminating the need for specific wake-up words and enabling natural, multi-interaction scenarios by using a type recognition model constructed based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional wake-up word based voice interaction is used, then the device can respond to voice commands, but the interaction becomes cumbersome requiring repeated wake-up words and cannot achieve natural language interaction experience
Solution Approach 1:
The patent extracts the wake-up word detection function from the general voice interaction process. By separating wake-up word recognition as an independent preliminary step, the system can identify user interaction intentions without requiring the wake-up word to be present in every command, thereby simplifying the interaction flow while maintaining responsiveness
Solution Approach 2:
The system performs wake-up word recognition as a preliminary action before main voice command processing. This preliminary detection enables the system to prepare for interaction mode switching in advance, allowing subsequent voice commands to be processed more naturally without repeatedly triggering wake-up sequences
2Measurement precision
If simple voice features are used for interaction recognition, then the processing is simple, but the accuracy of recognizing user intentions is poor leading to incorrect interactions
Solution Approach 1:
The patent combines multiple acoustic features (MFCC, pitch, energy) into a composite feature set for voice interaction recognition. This composite approach integrates different characteristics of voice signals to improve recognition accuracy, similar to how composite materials combine different properties to achieve superior performance
Solution Approach 2:
The system introduces an intermediary processing layer that transforms raw acoustic features into interaction intention predictions. This intermediary model acts as a bridge between simple voice features and complex interaction decisions, enabling accurate intention recognition without directly exposing the complexity of the processing pipeline
3Reliability
If wake-up word based interaction is used, then the device can identify interaction triggers, but false wake-ups occur and reduce interaction reliability
Solution Approach 1:
The system implements feedback mechanisms where the recognition model continuously learns from interaction outcomes. By analyzing whether detected wake-up words lead to successful interactions or false triggers, the system adjusts its recognition thresholds and parameters to reduce false wake-ups while maintaining reliable interaction detection
Solution Approach 2:
The patent dynamically adjusts recognition parameters such as confidence thresholds and time windows based on contextual information. By changing these parameters adaptively, the system can distinguish between genuine wake-up intentions and environmental noise, thereby reducing false wake-ups while maintaining high reliability for legitimate interactions
Data Source
AI summary
A method, apparatus, device, and storage medium for voice interaction. A specific embodiment of the method includes: extracting an acoustic feature from received voice data, the acoustic feature indicating a short-term amplitude spectrum characteristic of the voice data; applying the acoustic feature to a type recognition model to determine an intention type of the voice data, the intention type being one of an interaction intention type and a non-interaction intention type, and the type recognition model being constructed based on the acoustic feature of training voice data; and performing an interaction operation indicated by the voice data, based on determining that the intention type is the interaction intention type.