Two-Stage Hotphrase Detection for Flexible Low-Power Voice Triggering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional low-powered hotphrase detection models lack flexibility, requiring precise command sequences and failing to accommodate variations in user phrasing, limiting the range of commands that can be recognized by always-on assistant-enabled devices.
Innovation Solution
Implement a two-stage hotphrase detection system where a first stage detects multiple hotwords in an utterance and aggregates them to identify a full hotphrase, followed by a second-stage validation using automated speech recognition to confirm the presence of the hotphrase, allowing for flexible command recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional low-powered hotphrase detection models are used, then power consumption is reduced, but the flexibility and range of recognized commands are limited
Solution Approach 1:
The system segments hotphrase detection into two distinct stages: a first stage using low-powered detection models to identify candidate hotphrases, and a second stage using more computationally intensive automated speech recognition to validate and confirm hotphrases. This segmentation allows the system to maintain low average power consumption while achieving high flexibility in command recognition, as the intensive processing is only applied when necessary in the second stage.
Solution Approach 2:
The first stage of hotphrase detection performs preliminary filtering by detecting candidate hotphrases using low-powered models before committing to full speech recognition processing. This preliminary action reduces the number of utterances that require intensive processing, thereby maintaining energy efficiency while preparing for flexible command recognition in the second stage.
2Device complexity
If a single-stage hotphrase detection model is used, then device complexity is reduced, but measurement precision and reliability of hotphrase identification are limited
Solution Approach 1:
The detection system is segmented into two functional stages: a first stage that identifies candidate hotphrases using simplified models, and a second stage that validates these candidates using more sophisticated automated speech recognition. This segmentation improves detection accuracy without proportionally increasing overall system complexity, as each stage can be optimized independently for its specific function.
Solution Approach 2:
The first-stage hotphrase detection model acts as an intermediary between the audio input and the second-stage automated speech recognition. It filters and prepares candidate hotphrases, reducing the burden on the second stage and improving overall detection precision while maintaining manageable system complexity through clear functional separation.
Data Source
AI summary
A method includes receiving audio data corresponding to an utterance spoken by the user and captured by the user device. The utterance includes a command for a digital assistant to perform an operation. The method also includes determining, using a hotphrase detector configured to detect each trigger word in a set of trigger words associated with a hotphrase, whether any of the trigger words in the set of trigger words are detected in the audio data during the corresponding fixed-duration time window. The method also includes determining identifying, in the audio corresponding to the utterance, the hotphrase when each other trigger word in the set of trigger words was also detected in the audio data. The method also includes triggering an automated speech recognizer to perform speech recognition on the audio data when the hotphrase is identified in the audio data corresponding to the utterance.