Adaptive Filter Zone Segmentation for Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition technologies face challenges in accurately distinguishing and recognizing speech in complex sound environments, such as multiple speakers talking simultaneously, leading to incorrect recognition of user intentions.
Innovation Solution
The method involves collecting audio signals from multiple sound zones, determining the presence of key speech based on sound energy, adjusting the step size of adaptive filters in each zone, and performing adaptive filtering to enhance the key speech while suppressing non-critical sounds, thereby improving speech recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If speech recognition is performed on all audio signals from multiple sound zones, then comprehensiveness of recognition is improved, but recognition accuracy deteriorates due to interference from non-key speech
Solution Approach 1:
The patent divides the audio processing space into multiple sound zones with separate adaptive filters. Each filter independently processes audio signals from its corresponding sound zone, segmenting the overall recognition task into zone-specific subtasks. This allows comprehensive coverage of all zones while maintaining accuracy within each zone by focusing computational resources locally.
Solution Approach 2:
The patent applies different step sizes to adaptive filters in different sound zones based on local conditions. The step size of each adaptive filter is adjusted according to the determined result of its corresponding sound zone, allowing each zone to have optimized filtering parameters suited to its specific acoustic environment and speech characteristics.
2Device complexity
If adaptive filtering is applied to all sound zones with uniform parameters, then processing simplicity is maintained, but speech recognition accuracy deteriorates in complex sound environments
Solution Approach 1:
The patent makes the adaptive filter parameters dynamic by adjusting the step size of each adaptive filter according to the determined result of its corresponding sound zone. This dynamic adaptation allows the system to respond to changing acoustic conditions in each zone, improving recognition accuracy in complex environments while maintaining a relatively simple overall architecture.
Solution Approach 2:
The patent changes the step size parameter of adaptive filters based on the determined presence of key speech in each sound zone. When key speech is detected, the step size is adjusted to optimize convergence and filtering performance for that specific zone, thereby improving accuracy without requiring complete redesign of the filtering system.
3Speed
If the step size of adaptive filters is increased to speed up convergence, then processing speed is improved, but filtering precision deteriorates
Solution Approach 1:
The patent applies different step sizes to different adaptive filters based on local sound zone characteristics. Each sound zone receives a step size optimized for its specific conditions, allowing fast convergence in zones where it is appropriate while maintaining high filtering precision in zones where accuracy is critical.
Solution Approach 2:
The patent adjusts the step size parameter dynamically based on the determined result of each sound zone. This parameter adaptation allows the system to optimize the trade-off between convergence speed and filtering precision for each individual zone, rather than using a fixed step size that must compromise between these conflicting requirements globally.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Embodiments of the present disclosure provide a speech recognition method, a speech recognition apparatus, and a medium, and relate to a field of artificial intelligence speech recognition technologies. The method includes: obtaining (S101) audio signals collected by microphones in at least two sound zones; determining (S102) whether each audio signal includes a key speech according to sound energy of the audio signal to acquire a determined result; adjusting (S103) an adaptive adjustment parameter of an adaptive filter in each sound zone according to the determined result; controlling (S104) the adaptive filter to perform adaptive filtering processing on the audio signal collected in the sound zone corresponding to the adaptive filter according to the adaptive adjustment parameter, and outputting a filtered signal; and performing (S105) speech recognition on the filtered signal.