Mobile Speech Recognition Noise Filtering via Wireless Beacons
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile devices face challenges in accurately performing speech recognition in noisy environments due to interference from audio noise sources like industrial machinery, radios, and vehicle noise, which affects the proximity and characteristics of the noise.
Innovation Solution
The implementation of intelligent audio noise filtering in mobile devices using wireless communication signals, such as Bluetooth Low-Energy beacons, to determine audio noise profiles and proximity, allowing for effective filtering of noise and enhancement of speech recognition processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If audio noise filtering is applied to improve speech recognition accuracy, then speech recognition accuracy is improved, but device complexity increases
Solution Approach 1:
The audio signal is segmented into speech components and noise components using spectral subtraction and adaptive filtering. The noise profile is separately captured and stored, then applied to filter the incoming audio stream, separating the desired speech from the background noise without requiring complex real-time analysis of the entire audio spectrum.
Solution Approach 2:
The noise profile is captured and stored in advance during periods when no speech is present. This preliminary characterization of the noise environment allows the system to apply pre-computed filtering parameters during speech recognition, avoiding the need for complex real-time noise analysis and reducing computational complexity during the actual speech processing phase.
2Measurement precision
If wireless communication signals are used to determine noise profiles and proximity, then speech recognition accuracy is improved, but use of energy increases
Solution Approach 1:
The noise profile determination and proximity calculation functions are extracted from the main speech recognition processing path and implemented as separate, periodically-executing routines. The wireless communication signals are captured and processed independently to build noise profiles, which are then reused multiple times during speech recognition without requiring continuous signal processing, thereby reducing overall energy consumption.
Solution Approach 2:
The system performs noise profile capture and proximity determination periodically rather than continuously. Wireless signals are monitored at intervals to update noise profiles, and these profiles are applied to multiple speech recognition tasks between updates. This periodic approach maintains accurate noise characterization while significantly reducing the energy required compared to continuous monitoring and processing.
3Measurement precision
If audio noise filtering is implemented, then speech recognition accuracy is improved, but loss of information increases
Solution Approach 1:
The filtering operation is applied selectively in the frequency domain rather than uniformly across all frequencies. The spectral subtraction algorithm identifies and removes noise components only in frequency bins where noise is dominant, while preserving speech components in frequency regions where speech energy is concentrated. This localized filtering approach maintains speech information while removing noise.
Solution Approach 2:
The system uses the captured noise profile as feedback to continuously adjust the filtering parameters during speech recognition. The adaptive filter iteratively refines its noise estimation by comparing the filtered output with the original signal, and adjusts its parameters accordingly. This feedback mechanism ensures that speech information is preserved while noise is removed, as the filter learns to distinguish between speech and noise components based on the recorded noise characteristics.
Data Source
AI summary
A method improves speech recognition using a device located in proximity to a machine emitting high levels of audio noise. The microphone of the device receives the audio noise emitted by the machine and the speech emitted by a user and generates a composite signal. The device also receives a wireless communication signal from the machine comprising information on an audio noise profile and the proximity of the machine relative to the device. The audio noise profile is a representation of the audio noise emitted by the machine. Based on this information, the device determines a filter for filtering the composite signal to mitigate the audio noise before initiating the speech recognition process. The method improves speech recognition in a high audio noise environment.