Variable-Time Smoothing for Steady-State Noise Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Variable acoustic noise in vehicles degrades the quality of music or speech, making it difficult to distinguish soft sounds and reducing the fidelity of music or intelligibility of speech, which existing technologies have not effectively addressed.
Innovation Solution
A method and system that use adaptive time and frequency smoothing to estimate noise floors in audio processing systems, adjusting smoothing parameters based on speech activity and averaging noise estimates across multiple frames and frequency bins to improve noise reduction and speech recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed time smoothing is used for noise estimation, then processing is simple, but noise estimation accuracy is insufficient under variable acoustic conditions
Solution Approach 1:
The patent applies dynamics by making the smoothing parameter time-variant rather than fixed. The parameter is adjusted based on speech activity detection and noise characteristics, allowing the system to adapt to changing acoustic environments in real-time. This resolves the contradiction by improving measurement precision through adaptive smoothing while managing complexity through rule-based parameter adjustment.
Solution Approach 2:
The patent changes the smoothing parameter based on detected speech presence and noise conditions. When speech is detected, the parameter is modified to reduce smoothing strength, while during silence, stronger smoothing is applied. This parameter adaptation enables accurate noise estimation across variable conditions without requiring an overly complex processing system.
2Measurement precision
If strong smoothing is applied to reduce noise, then noise floor estimation improves, but speech distortions increase
Solution Approach 1:
The system dynamically adjusts smoothing strength based on speech activity detection. During speech segments, smoothing is reduced to preserve speech information, while during non-speech segments, smoothing is strengthened to accurately estimate noise floor. This dynamic adaptation resolves the contradiction between noise reduction and speech preservation.
Solution Approach 2:
The patent employs periodic evaluation of speech presence to modulate smoothing parameters. The system continuously monitors for speech and switches between strong and weak smoothing modes in response to periodic speech-on/off patterns, enabling accurate noise estimation during silent periods while preserving speech during active periods.
3Measurement precision
If smoothing parameter is adjusted frequently to adapt to changing conditions, then noise estimation accuracy improves, but processing time increases
Solution Approach 1:
The system uses feedback from speech activity detection to adjust smoothing parameters. The feedback mechanism monitors speech presence and automatically modulates the parameter, avoiding unnecessary processing during stable conditions. This feedback-based approach improves estimation accuracy when needed while minimizing processing time during steady-state conditions.
Solution Approach 2:
The patent changes parameters only when necessary based on detected condition changes (speech onset/offset, noise level transitions). Rather than continuous adjustment, the system applies discrete parameter changes triggered by significant condition changes, reducing processing overhead while maintaining adaptive accuracy.
4Reliability
If frequency domain processing is applied to smooth noise estimates, then noise reduction performance improves, but computational complexity increases
Solution Approach 1:
The patent segments the frequency spectrum into multiple bins and applies different smoothing strategies to different frequency regions. This segmentation allows targeted processing that improves noise reduction in relevant frequency bands while reducing unnecessary computation in less critical regions, managing complexity through selective processing.
Solution Approach 2:
The system applies different smoothing characteristics to different frequency bins based on local noise characteristics and speech content. This local quality approach enables optimized noise reduction in specific frequency ranges while avoiding excessive processing elsewhere, improving overall reliability without uniformly increasing computational complexity across the entire spectrum.
Data Source
AI summary
A method includes receiving multiple frames of time-domain data that includes noise, and computing, for a first frame of the multiple frames, a frequency domain value for each of multiple frequency bins, each frequency bin representing a corresponding range of frequencies. The method also includes determining that a first frequency domain value corresponding to a first frequency bin is less than or equal to a first threshold value, and in response, updating the first frequency domain value based on a function of (i) a smoothing parameter, and (ii) a second frequency domain value corresponding to the first frequency bin. The second frequency domain value is computed using one or more preceding frames of the multiple frames. The method further includes determining a noise floor corresponding to the first frequency bin using the updated first frequency domain value.