Acoustic Signal Separation Using Environment-Weighted Mask Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional signal processing techniques require complicated preparation processing for creating auxiliary information regarding a target sound source and may lower the performance of separating multiple acoustic signals from a mixed acoustic signal, especially when noise is present.
Innovation Solution
A signal processing device that acquires a mixed acoustic signal, converts it into a mixed feature amount, estimates masks using an environment sound feature amount, and separates the signals without prior auxiliary information, utilizing multiple acoustic models for feature conversion and extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If auxiliary information regarding a target sound source is created in advance, then the performance of separating acoustic signals is improved, but the preparation processing becomes complicated and time-consuming
Solution Approach 1:
The system performs preliminary extraction of environment sound features from the mixed acoustic signal before mask estimation. The environment sound feature amount is extracted in advance and stored, then used during the actual signal separation process to guide mask estimation without requiring complex manual preparation of auxiliary information about target sound sources.
Solution Approach 2:
The system uses the mixed acoustic signal itself to extract environment sound features, eliminating the need for external auxiliary information sources. The environment sound is automatically identified and extracted from the mixed signal, and this extracted information is then used to improve the separation of target sound sources, making the system self-sufficient.
2Measurement precision
If conventional methods are used without environment sound consideration, then processing is simpler, but signal separation performance deteriorates in noisy environments
Solution Approach 1:
The system extracts the environment sound component from the mixed acoustic signal separately. By identifying and extracting the environment sound feature amount from the mixed signal, the system isolates the noise component, allowing it to be used as reference information for improving the separation of target sound sources in noisy environments.
Solution Approach 2:
The environment sound feature amount acts as an intermediary between the mixed acoustic signal and the mask estimation process. This intermediate representation captures the noise characteristics and translates them into useful information that guides the mask estimation, thereby improving separation performance without directly modifying the target sound source extraction.
3Measurement precision
If auxiliary information is required for signal separation, then accuracy is improved, but real-time processing capability is reduced
Solution Approach 1:
The environment sound feature amount is extracted and prepared in advance from the mixed acoustic signal. This preliminary extraction creates ready-to-use auxiliary information that can be quickly utilized during real-time mask estimation, eliminating the need for complex on-the-fly processing while maintaining high separation accuracy.
Data Source
AI summary
A signal processing device includes: a mixed feature amount conversion part that converts a mixed acoustic signal including a plurality of acoustic signals into a mixed feature amount; a mask estimation part that estimates a plurality of masks based on the mixed feature amount; an acoustic signal conversion part that converts a plurality of separated feature amounts calculated using the plurality of masks into a plurality of separated acoustic signals; an environment acoustic signal extraction part that extracts, as an environment acoustic signal, the mixed acoustic signal in the environment sound section estimated based on the plurality of separated acoustic signals from the mixed acoustic signal; and an environment sound feature amount conversion part that converts the environment acoustic signal into an environment sound feature amount. The mask estimation part estimates the plurality of masks based on the mixed feature amount weighted using the environment sound feature amount.