Speaker Output Suppression for Barge-In Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face challenges in accurately recognizing user input when it overlaps with audio output from text-to-speech systems, media players, or navigation systems, as the microphone captures both signals, leading to interference and reduced recognition accuracy.
Innovation Solution
A computer-implemented method and system that utilize a model trained to identify the output of a speaker device, allowing for the suppression of the audio output from the speaker device when user input is detected, using techniques such as i-vector based models or neural networks to differentiate between user voice and device output, thereby reducing the audio output level and improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the microphone captures both speaker device output and user input simultaneously, then the system can maintain continuous audio monitoring, but speech recognition accuracy deteriorates due to signal interference
Solution Approach 1:
The audio signal is segmented into different components (speaker device output and user input) using a trained model that can identify and separate these signals. The system divides the mixed audio stream into distinct sources, allowing accurate recognition of user speech even when captured simultaneously with speaker output.
Solution Approach 2:
A trained model acts as an intermediary between the mixed audio signal and the speech recognition system. This model processes the combined signal first, identifying and isolating user input from speaker output, before passing the separated user speech to the recognition engine, thereby eliminating interference.
2Measurement precision
If the audio output level of the speaker device is reduced when user input is detected, then speech recognition accuracy improves, but the user experience may deteriorate due to sudden audio changes
Solution Approach 1:
The system performs preliminary detection of user input using the trained model before actually reducing the speaker output level. By identifying user speech intent in advance and preparing for the audio level change, the system can smoothly transition the audio output rather than making abrupt changes, maintaining user experience while improving recognition accuracy.
3Adaptability or versatility
If a trained model is used to identify speaker device output, then the system can differentiate between user voice and device output, but device complexity increases
Solution Approach 1:
Instead of implementing complex real-time signal separation algorithms, the system creates a trained model that learns and copies the characteristics of speaker device output. This model serves as a reference representation that can be efficiently compared against incoming audio signals, providing accurate differentiation without requiring complex processing infrastructure.
Data Source
AI summary
The technology described in this document can be embodied in a computer-implemented method that includes receiving, at a processing system, a first signal including an output of a speaker device and an additional audio signal. The method also includes determining, by the processing system, based at least in part on a model trained to identify the output of the speaker device, that the additional audio signal corresponds to an utterance of a user. The method further includes initiating a reduction in an audio output level of the speaker device based on determining that the additional audio signal corresponds to the utterance of the user.