ESSV Konferenz Elektronische Sprachsignalverarbeitung

Title: Reduction of Aircraft Noise in UAV-Based Speech Signal Recordings by Quantile Based Noise Estimation

Authors: Enrico Lösch, Oliver Jokisch, Alexander Leipnitz, Ingo Siegert

Abstract:

In this article we survey, whether the signal-to-noise ratio of speech signals, superimposed by flight noise, can be significantly improved by advanced noise-estimation methods, such as Quantile Based Noise Estimation (QBNE) or Adaptive Quantile Based Noise Estimation (AQBNE) and a spectral subtraction of the respective noise components. Our test object, a typical commercial UAV (DJI Mavic Pro), has been extended by a lightweight 8-microphone array (vicDIVA). The speech recordings in a free-field environment are processed with a batterypowered Raspberry Pi 3, attached to the UAV. The source of the speech signals is located on the ground. During our recordings, the UAV is hovering over the sound source. In addition to the noise-estimation methods QBNE and AQBNE, different distances between sound source and UAV as well as the directional effects of the microphone array (beamforming and steering) are investigated.


Year: 2020
In session: Poster
Pages: 149 to 156