DeepEar Dataset

Dataset and demo code of DeepEar.

Training and Testing data

(Download) here.

Files

• DeepEar.py: Demo code of DeepEar.
  • TensorFlow 2.5.0
  • mat73 is required.
• DeepEar_weights.h5: Pretrained model weights.
• TrainData.mat: Extracted features for training data.
  • Matlab v7.3 version file.
  • Four column are gammatone coefficients of left ear, right ear, cross-correlation, and ground truth labels. Please refer to our paper for more details.
• TestData.mat: Extracted features for testing data.

Label format

1 X 56 vector. Please refer to our paper for more details.

        [1]: binary sectors. [2]: AoA (0~1). [3-7]: one-hot distance
        [8]: binary sectors. [9]: AoA (0~1). [10-14]: one-hot distance
        [15]: binary sectors. [16]: AoA (0~1). [17-21]: one-hot distance
        [22]: binary sectors. [23]: AoA (0~1). [24-28]: one-hot distance
        [29]: binary sectors. [30]: AoA (0~1). [31-35]: one-hot distance
        [36]: binary sectors. [37]: AoA (0~1). [38-42]: one-hot distance
        [43]: binary sectors. [44]: AoA (0~1). [45-49]: one-hot distance
        [50]: binary sectors. [51]: AoA (0~1). [52-56]: one-hot distance

Credits and License

The dataset and code are provided by Qiang Yang under the guidance of Prof. Yuanqing Zheng of The Hong Kong Polytechnic University (PolyU). They are licensed under CC-BY-NC.

For any questions, you may contact: qiang {dot} yang {at} connect {dot} polyu {dot} hk.

Cite this work

@inproceedings{yang22DeepEar,
title={DeepEar: Sound Localization with Binaural Microphones},
author={Yang, Qiang and Zheng, Yuanqing},
booktitle = {Proceedings of the International Conference on Computer Communications (INFOCOM~'22)},
year = {2022},
publisher={IEEE}
}