Pre-computed Hotword Models for Mobile Speech Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems struggle to efficiently detect hotwords in a multi-user environment without requiring transcription or semantic interpretation of the spoken words.
Innovation Solution
The system provides pre-computed hotword models to mobile computing devices, allowing them to detect candidate hotwords based on acoustic features without the need for transcription or semantic interpretation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech recognition systems transcribe and semantically interpret spoken words to detect hotwords, then detection accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The system pre-computes and stores hotword models during an offline training phase, so that during runtime, the mobile device can directly compare acoustic features against these pre-computed models without performing transcription or semantic interpretation. This preliminary action separates the computationally intensive model training from the real-time detection process.
Solution Approach 2:
The system segments the hotword detection process into two distinct phases: (1) offline model computation and storage, and (2) online acoustic feature comparison. This segmentation allows the complex transcription and interpretation tasks to be performed only once during model creation, while runtime detection uses simple acoustic matching.
2Adaptability or versatility
If the system processes all utterances through transcription and semantic interpretation, then comprehensive understanding is achieved, but energy consumption increases
Solution Approach 1:
The system performs only the necessary minimum action for hotword detection by comparing acoustic features directly against pre-computed hotword models, without performing full transcription or semantic interpretation. This partial action approach processes only the specific acoustic patterns needed for hotword recognition, conserving energy while maintaining detection capability.
3Reliability
If the system uses complex transcription and semantic interpretation for hotword detection, then detection reliability improves, but device complexity increases
Solution Approach 1:
The system extracts only the essential acoustic feature comparison function from the full speech recognition pipeline, separating hotword detection from transcription and semantic interpretation. This extraction allows the mobile device to perform simple acoustic matching without implementing the complex transcription and interpretation components.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for obtaining, for each of multiple words or sub-words, audio data corresponding to multiple users speaking the word or sub-word; training, for each of the multiple words or sub-words, a pre-computed hotword model for the word or sub-word based on the audio data for the word or sub-word; receiving a candidate hotword from a computing device; identifying one or more pre-computed hotword models that correspond to the candidate hotword; and providing the identified, pre-computed hotword models to the computing device.