DroneAudioNet: Noise Suppression for Drone Audition-based Search and Rescue

author list hidden in review stage

submitted to Neurips 2026

Abstract

Microphones mounted on UAVs enable aerial acoustic scene analysis applications such as search-and-rescue, wildlife monitoring, and industrial inspection. However, drone rotor noise often dominates the mixture signal at SNRs well below −10 dB, making source recovery extremely challenging. Existing enhancement and source separation methods are typically designed for near-balanced mixtures and degrade substantially in drone audition settings. In this work, we propose DRONEAUDIONET, a drone noise suppression method that reframes a source separation model as a drone noise estimator. To better model drone-dominant mixtures, we introduce a learnable mask-scaling mechanism that allows mask magnitudes beyond unity, together with an additive residual correction term for improved drone estimation and source recovery. We train and evaluate our model on a publicly available drone audition dataset and test generalizability on an out-of-domain dataset with unseen drone hardware and flight modes. Results show that DRONEAUDIONET consistently improves downstream sound classification performance, with the largest gains observed for human vocal sounds. Our findings demonstrate the importance of drone-specific modeling for robust aerial acoustic perception and highlight the potential of source separation methods for real-world drone-assisted search-and-rescue.

DroneAudioNet teaser

Examples

This section contains four comparison tables. Three summarize in-domain examples at input SNRs from −10 dB to 0 dB, from −20 dB to −10 dB, and from −30 dB to −20 dB. The fourth reports an out-of-domain comparison on the Dregon dataset. Each cell includes the audio clip and a spectrogram for the same clip.

Input SNR: -10 dB to 0 dB
Example noisy uss zeroshot audiosep audiosep-finetuned droneaudionet
crying
Input SNR: -10 dB to 0 dB crying noisy spectrogram
Input SNR: -10 dB to 0 dB crying uss spectrogram
Input SNR: -10 dB to 0 dB crying zeroshot spectrogram
Input SNR: -10 dB to 0 dB crying audiosep spectrogram
Input SNR: -10 dB to 0 dB crying audiosep-finetuned spectrogram
Input SNR: -10 dB to 0 dB crying droneaudionet spectrogram
female
Input SNR: -10 dB to 0 dB female noisy spectrogram
Input SNR: -10 dB to 0 dB female uss spectrogram
Input SNR: -10 dB to 0 dB female zeroshot spectrogram
Input SNR: -10 dB to 0 dB female audiosep spectrogram
Input SNR: -10 dB to 0 dB female audiosep-finetuned spectrogram
Input SNR: -10 dB to 0 dB female droneaudionet spectrogram
Input SNR: -20 dB to -10 dB
Example noisy uss zeroshot audiosep audiosep-finetuned droneaudionet
crying
Input SNR: -20 dB to -10 dB crying noisy spectrogram
Input SNR: -20 dB to -10 dB crying uss spectrogram
Input SNR: -20 dB to -10 dB crying zeroshot spectrogram
Input SNR: -20 dB to -10 dB crying audiosep spectrogram
Input SNR: -20 dB to -10 dB crying audiosep-finetuned spectrogram
Input SNR: -20 dB to -10 dB crying droneaudionet spectrogram
Input SNR: -30 dB to -20 dB
Example noisy uss zeroshot audiosep audiosep-finetuned droneaudionet
crying
Input SNR: -30 dB to -20 dB crying noisy spectrogram
Input SNR: -30 dB to -20 dB crying uss spectrogram
Input SNR: -30 dB to -20 dB crying zeroshot spectrogram
Input SNR: -30 dB to -20 dB crying audiosep spectrogram
Input SNR: -30 dB to -20 dB crying audiosep-finetuned spectrogram
Input SNR: -30 dB to -20 dB crying droneaudionet spectrogram
Out-of-domain (Dregon)
Dregon dataset
Method male female crying human sounds non human sounds
original
Dregon male original spectrogram
Dregon female original spectrogram
Dregon crying original spectrogram
Dregon human sounds original spectrogram
Dregon non human sounds original spectrogram
AudioSep
Dregon male AudioSep spectrogram
Dregon female AudioSep spectrogram
Dregon crying AudioSep spectrogram
Dregon human sounds AudioSep spectrogram
Dregon non human sounds AudioSep spectrogram
audiosep-finetuned
Dregon male audiosep-finetuned spectrogram
Dregon female audiosep-finetuned spectrogram
Dregon crying audiosep-finetuned spectrogram
Dregon human sounds audiosep-finetuned spectrogram
Dregon non human sounds audiosep-finetuned spectrogram
DroneAudioNet
Dregon male DroneAudioNet spectrogram
Dregon female DroneAudioNet spectrogram
Dregon crying DroneAudioNet spectrogram
Dregon human sounds DroneAudioNet spectrogram
Dregon non human sounds DroneAudioNet spectrogram

Download

Citations

BibTeX will be provided after acceptance / camera-ready.

@inproceedings{droneaudionet2026,
  title     = {DroneAudioNet: Noise Suppression for Drone Audition-based Search and Rescue},
  author    = {Anonymous},
  booktitle = {Advances in Neural Information Processing Systems},
  year      = {2026}
}