Language-Independent Wake-Up Word Detection via Acoustic Feature Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mobile devices operating in low-power mode lack the capability for language-specific wake-up word detection due to the unavailability of trained acoustic models for the user's language, necessitating manual triggers that are inconvenient or impossible in certain situations.
Innovation Solution
Utilize an acoustic model trained in a source language to recognize wake-up words in a target language by deriving acoustic features from enrollment utterances, adapting the model to the user's speech, and storing reference sequences for language-independent wake-up word detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the device uses an acoustic model trained in a source language to detect wake-up words in a target language, then language versatility is improved, but measurement precision may deteriorate
Solution Approach 1:
The patent uses acoustic features as an intermediary representation that is language-independent. The acoustic model processes these features to generate speech units, which are then compared to reference sequences. This intermediary approach allows the system to work across different languages without requiring language-specific acoustic models, resolving the contradiction between versatility and precision.
Solution Approach 2:
The system changes the parameter space by transforming raw acoustic input into acoustic features, then into speech units, and finally comparing against reference sequences. This parameter transformation enables language-independent processing while maintaining detection accuracy through the structured progression of feature extraction and comparison.
2Use of energy by moving object
If the device performs wake-up word detection locally without network connections, then power consumption is reduced, but device complexity increases
Solution Approach 1:
The patent segments the speech recognition task into two parts: wake-up word detection (performed locally on the device) and full speech recognition (performed online when needed). This segmentation allows the device to maintain simple low-power operation for wake-up detection while reserving complex processing for when the device is already active and connected.
Solution Approach 2:
The system performs only partial speech processing locally - specifically, wake-up word detection using acoustic features and speech unit comparison. Full speech recognition is left for online processing. This partial action approach enables local operation with reduced power consumption while avoiding the need for complete speech recognition capabilities on the device.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A user device configured to perform wake-up word detection in a target language. The user device comprises at least one microphone (430) configured to obtain acoustic information from the environment of the user device, at least one computer readable medium (435) storing an acoustic model (150) trained on a corpus of training data (105) in a source language different than the target language, and storing a first sequence of speech units obtained by providing acoustic features (110) derived from audio comprising the user speaking a wake-up word in the target language to the acoustic model (150), and at least one processor (415,425) coupled to the at least one computer readable medium (435) and programmed to perform receiving, from the at least one microphone (430), acoustic input from the user speaking in the target language while the user device is operating in a low-power mode, applying acoustic features derived from the acoustic input to the acoustic model (150) to obtain a second sequence of speech units corresponding to the acoustic input, determining if the user spoke the wake-up word at least in part by comparing the first sequence of speech units to the second sequence of speech units, and exiting the low-power mode if it is determined that the user spoke the wake-up word.