MAP estimation-based noise suppression and blind source separation using single voice activity detection

2014 Fiscal Year Final Research Report

• Project/Area Number: 24500204
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Osaka University
• Principal Investigator: KAWAMURA Arata 大阪大学, 基礎工学研究科, 准教授 (60362646)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 単一話者区間 / 音源分離 / ノイズ除去 / 事後確率最大化 / 統計処理

Outline of Final Research Achievements

A blind source separation method using two microphones has been investigated. In a practical situation, environmental noise and reverberation exist. Since they degrade source separation quality, we have to remove such undesired effects. To establish an effective blind source separation method, we employ a single voice activity detector which detects single talk segments. These segments give the target source locations. Addition to it, a MAP (maximum a posteriori) estimation-based noise suppressor is introduced as a post-processor for improving the speech quality of the separated signals. Test speech signals are transmitted from loudspeakers and captured at a stereo microphone in a practical reverberant environment. Simulation results showed that the observed speech signals are effectively separated by the proposed method.

Free Research Field

音声信号処理