Device Arbitration for Speech Processing Using ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In environments with multiple speech processing devices, determining which device is closest to the user to process spoken commands accurately is challenging due to varying microphone sensitivities and acoustic conditions, especially in reverberant rooms with noise interference, and existing methods like TDOA algorithms require significant resources and synchronization.
Innovation Solution
An end-to-end machine learning-based device arbitration system using deep neural networks (DNNs) for feature extraction and classification to predict the closest device to the speaker, which generates feature data representing audio characteristics and combines them to make arbitration decisions without requiring synchronization or large memory resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If TDOA algorithms are used to determine the closest device, then device arbitration can be performed, but significant computational resources and synchronization are required
Solution Approach 1:
The patent replaces complex signal processing algorithms (TDOA) with a simpler machine learning classification system. The DNN model processes audio features directly to predict the closest device, eliminating the need for computationally intensive time difference calculations and inter-device synchronization mechanisms.
Solution Approach 2:
The patent introduces an intermediary machine learning model that sits between audio capture and device selection. This model processes audio features and outputs a prediction of the closest device, acting as a mediator that simplifies the arbitration process while maintaining accuracy.
2Measurement precision
If TDOA algorithms are used to determine the closest device, then device arbitration can be performed, but synchronization between devices is required
Solution Approach 1:
The patent replaces synchronization-dependent TDOA algorithms with an independent machine learning classification system. Each device processes its own audio features through the DNN model, and predictions are aggregated without requiring temporal synchronization between devices.
3Reliability
If multiple devices capture audio data, then speech processing can be performed, but resource usage increases and latency occurs
Solution Approach 1:
The patent extracts and processes only the necessary audio features (energy, spectral characteristics) rather than capturing and processing entire audio streams from all devices. This selective feature extraction reduces resource consumption while maintaining the ability to accurately identify the closest device.
Solution Approach 2:
The patent transforms raw audio data into compressed feature representations (energy, spectral features) that capture essential information while reducing data volume. This parameter transformation enables efficient processing with lower computational resources and reduced latency.
Data Source
AI summary
Devices and techniques are generally described for nearest device arbitration. In various examples, a first device may receive first audio data representing a wakeword spoken by a first speaker at a first time. In some examples, a second device may receive second audio data representing the wakeword spoken by the first speaker at the first time. In some cases, the first device may generate first feature data representing the first audio data and the second device may generate second feature data representing the second audio data. In various examples, a machine learning model may use the first feature data and the second feature data to generate first prediction data representing a prediction that the first device is closer to the first speaker than the second device.