Sound source isolating device

The sound source separation device improves accuracy by using a neural network with group sparseness constraints to generate masks for monaural acoustic signals, addressing overfitting issues and enhancing separation performance.

WO2026126309A1 Publication Date: 2026-06-18NT T INC
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
NT T INC
Filing Date
2024-12-10
Publication Date
2026-06-18

AI Technical Summary

⚠Technical Problem

Existing sound source separation methods using neural networks, such as U-Net-based approaches, suffer from overfitting when encountering unknown acoustic signals, reducing accuracy in separating sound sources from monaural acoustic signals.

⚗Method used

A sound source separation device that incorporates a time-frequency domain signal generation unit, a mask generation unit with a group sparseness constraint specification, and a time-domain signal generation unit, utilizing a neural network to generate masks based on group sparseness constraints, thereby improving accuracy even with unknown acoustic signals.

🎯Benefits of technology

Enhances the accuracy of sound source separation by suppressing overfitting through the use of group sparseness constraints, ensuring effective separation of target and noise components in monaural acoustic signals.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

A sound source isolating device according to the present invention comprises a time-frequency domain signal generation unit, a mask generation unit, and a group sparse constraint designation unit. The time-frequency domain signal generation unit generates a time-frequency domain signal from a time domain monaural acoustic signal in which stationary noise is mixed. The mask generation unit generates a time-frequency domain mask that isolates noise components in the time-frequency domain signal by solving an optimization problem that uses the time-frequency domain signal. The group sparse constraint designation unit designates, on the basis of acoustic features of the noise, a group sparse constraint for determining a loss term to be introduced into a loss function for solving an optimization problem.
Need to check novelty before this filing date? Find Prior Art