Speech Detection Without Wake Words
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech processing systems face inefficiencies due to the need for significant computing resources to process automatic speech recognition (ASR) and natural language understanding (NLU) within a reasonable time frame, often requiring distributed computing environments and relying on wake words to initiate processing, which can lead to unnecessary resource usage when no command is being given.
Innovation Solution
The system employs techniques to determine if incoming audio data includes system-directed speech without a detected wake command, using voice activity detection and other methods to filter out non-relevant audio, allowing continuous audio capture and processing even without a wake word, and enabling devices to send audio data to servers for further processing without initial wake word detection in certain circumstances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the system continuously processes all incoming audio data through ASR and NLU, then no command is missed, but computing resource usage increases significantly
Solution Approach 1:
The patent extracts and processes only relevant audio segments through voice activity detection and wake word detection, separating useful command-containing audio from irrelevant continuous audio background, thereby reducing computing resource consumption while maintaining command detection capability
Solution Approach 2:
The system performs preliminary filtering of audio data using voice activity detection and wake word detection before submitting to full ASR and NLU processing, preparing the audio data in advance to identify only those segments warranting intensive processing
2Use of energy by moving object
If the system relies on wake word detection to initiate processing, then computing resources are saved, but commands without wake words are missed
Solution Approach 1:
The system applies partial processing (voice activity detection and wake word detection) to all audio continuously, and excessive processing (full ASR and NLU) only to segments that pass the preliminary filters, achieving a balance between resource efficiency and command detection completeness
Solution Approach 2:
Preliminary voice activity and wake word detection is performed on all incoming audio before full processing, preparing segments in advance to determine which warrant intensive ASR and NLU processing
3Speed
If the system processes audio in real-time, then response speed is fast, but computing resources are consumed heavily
Solution Approach 1:
The audio processing pipeline is segmented into multiple stages: continuous voice activity detection, wake word detection, and conditional full ASR/NLU processing. This segmentation allows real-time processing of all audio through lightweight detectors while reserving intensive processing only for relevant segments
Data Source
AI summary
A speech-processing system capable of receiving and processing audio data to determine if the audio data includes speech that was intended for the system. Non-system directed speech may be filtered out while system-directed speech may be selected for further processing. A system-directed speech detector may use a trained machine learning model (such as a deep neural network or the like) to process a feature vector representing a variety of characteristics of the incoming audio data, including the results of automatic speech recognition and/or other data. Using the feature vector the model may output an indicator as to whether the speech is system-directed. The system may also incorporate other filters such as voice activity detection prior to speech recognition, or the like.