Background Noise Estimation for Audio Signal Music Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio coding technologies face challenges in accurately distinguishing between active speech/music and background noise, especially in low SNR conditions, leading to potential clipping and inefficiencies in discontinuous transmission systems.
Innovation Solution
An improved method for estimating background noise in audio signals, which reduces the current noise estimate when music is detected and the energy difference is below a threshold, without relying on feedback from activity detectors, allowing for more accurate sound activity detection and reducing the risk of speech/music being misclassified as background noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the activity detector is too efficient in detecting non-activity, then DTX efficiency is improved, but clipping occurs in active signals causing quality degradation
Solution Approach 1:
The patent changes the parameter of background noise estimation by introducing a music detection mechanism that identifies non-stationary signals. When music is detected, the system prevents updates to the background noise estimate, thereby avoiding parameter drift that would cause clipping. This resolves the contradiction by adapting the estimation behavior based on signal type, maintaining both DTX efficiency and signal quality.
2Reliability
If the activity detector is not efficient enough, then clipping is avoided, but background noise segments are misclassified as active signals reducing DTX efficiency
Solution Approach 1:
The system dynamically changes the estimation parameter behavior based on music detection results. When music is detected through non-stationarity analysis, the background noise estimate is held constant, preventing false updates that would reduce detection accuracy. This maintains reliable signal quality while preserving DTX efficiency through accurate background modeling during stationary periods.
3Measurement precision
If background noise estimate is continuously updated, then tracking of non-stationary noise is improved, but music segments are misclassified as background noise
Solution Approach 1:
The patent introduces a feedback mechanism where the update process continuously monitors signal stationarity. When non-stationarity indicative of music is detected, the feedback loop prevents background noise estimate updates. This resolves the contradiction by using feedback to distinguish between transient noise variations that should be tracked and musical content that should preserve the existing estimate.
Solution Approach 2:
The system changes the estimation parameter update rule based on detected signal characteristics. During stationary periods, normal updating occurs to track noise accurately. During non-stationary periods identified as music, the update parameter is changed to prevent modifications. This adaptive parameter change maintains both tracking accuracy and classification reliability.
4Measurement precision
If decision feedback is used for background estimation, then estimation accuracy is improved, but complexity of the system increases
Solution Approach 1:
The patent changes the operational parameter of the estimation algorithm based on music detection. Instead of always using decision feedback or never using it, the system dynamically adjusts whether to apply feedback based on signal stationarity. When music is detected, feedback is disabled; during stationary speech, feedback is enabled. This reduces overall system complexity while maintaining high estimation accuracy when needed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method and an apparatus for background noise estimation in an audio signal segment. The method comprises determining if the audio signal segment comprises music, and reducing a current noise estimate of each sub-band where the current noise estimate exceeds a minimum value when the audio signal segment is determined to comprise music. This is to be performed when the difference between an energy of the audio signal segment and a long-term minimum energy is below a threshold value, but no pause is detected in the audio signal segment. Performing the method enables a more adequate sound activity detection based on the background noise estimate, as compared to the prior art.