TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Transfer Learning of Weakly Labelled Audio

Tutkimustuotosvertaisarvioitu

Standard

Transfer Learning of Weakly Labelled Audio. / Diment, Aleksandr; Virtanen, Tuomas.

IEEE Workshop on Applications of Signal Processing to Audio and Acoustics. 2017. s. 6-10 (IEEE Workshop on Applications of Signal Processing to Audio and Acoustics).

Tutkimustuotosvertaisarvioitu

Harvard

Diment, A & Virtanen, T 2017, Transfer Learning of Weakly Labelled Audio. julkaisussa IEEE Workshop on Applications of Signal Processing to Audio and Acoustics. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, Sivut 6-10, IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 1/01/00. https://doi.org/10.1109/WASPAA.2017.8169984

APA

Diment, A., & Virtanen, T. (2017). Transfer Learning of Weakly Labelled Audio. teoksessa IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (Sivut 6-10). (IEEE Workshop on Applications of Signal Processing to Audio and Acoustics). https://doi.org/10.1109/WASPAA.2017.8169984

Vancouver

Diment A, Virtanen T. Transfer Learning of Weakly Labelled Audio. julkaisussa IEEE Workshop on Applications of Signal Processing to Audio and Acoustics. 2017. s. 6-10. (IEEE Workshop on Applications of Signal Processing to Audio and Acoustics). https://doi.org/10.1109/WASPAA.2017.8169984

Author

Diment, Aleksandr ; Virtanen, Tuomas. / Transfer Learning of Weakly Labelled Audio. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics. 2017. Sivut 6-10 (IEEE Workshop on Applications of Signal Processing to Audio and Acoustics).

Bibtex - Lataa

@inproceedings{23e81fb936664e09b579514784518b72,
title = "Transfer Learning of Weakly Labelled Audio",
abstract = "Many machine learning tasks have been shown solvable with impressive levels of success given large amounts of training data and computational power. For the problems which lack data sufficient to achieve high performance, methods for transfer learning can be applied. These refer to performing the new task while having prior knowledge of the nature of the data, gained by first performing a different task, for which training data is abundant. Shown successful for other machine learning tasks, transfer learning is now investigated in audio analysis. We propose to solve the weakly labelled problem of sound event tagging with small amounts of training data by transferring the abstract knowledge about the nature of audio data from another tagging task. The proposed methods constitute pre-Training of a recurrent neural network or its parts to perform one tagging task given abundant and diverse training data, and then using it or its parts for a new task of tagging sound events of different nature, for which the data is limited. Several architectures for such transfer are proposed and evaluated, showing impressive classification accuracy of 83.4{\%} with gains of up to 20 percentage points over the baseline given as little as 36 training samples for the target task.",
author = "Aleksandr Diment and Tuomas Virtanen",
year = "2017",
doi = "10.1109/WASPAA.2017.8169984",
language = "English",
isbn = "978-1-5386-1631-4",
series = "IEEE Workshop on Applications of Signal Processing to Audio and Acoustics",
pages = "6--10",
booktitle = "IEEE Workshop on Applications of Signal Processing to Audio and Acoustics",

}

RIS (suitable for import to EndNote) - Lataa

TY - GEN

T1 - Transfer Learning of Weakly Labelled Audio

AU - Diment, Aleksandr

AU - Virtanen, Tuomas

PY - 2017

Y1 - 2017

N2 - Many machine learning tasks have been shown solvable with impressive levels of success given large amounts of training data and computational power. For the problems which lack data sufficient to achieve high performance, methods for transfer learning can be applied. These refer to performing the new task while having prior knowledge of the nature of the data, gained by first performing a different task, for which training data is abundant. Shown successful for other machine learning tasks, transfer learning is now investigated in audio analysis. We propose to solve the weakly labelled problem of sound event tagging with small amounts of training data by transferring the abstract knowledge about the nature of audio data from another tagging task. The proposed methods constitute pre-Training of a recurrent neural network or its parts to perform one tagging task given abundant and diverse training data, and then using it or its parts for a new task of tagging sound events of different nature, for which the data is limited. Several architectures for such transfer are proposed and evaluated, showing impressive classification accuracy of 83.4% with gains of up to 20 percentage points over the baseline given as little as 36 training samples for the target task.

AB - Many machine learning tasks have been shown solvable with impressive levels of success given large amounts of training data and computational power. For the problems which lack data sufficient to achieve high performance, methods for transfer learning can be applied. These refer to performing the new task while having prior knowledge of the nature of the data, gained by first performing a different task, for which training data is abundant. Shown successful for other machine learning tasks, transfer learning is now investigated in audio analysis. We propose to solve the weakly labelled problem of sound event tagging with small amounts of training data by transferring the abstract knowledge about the nature of audio data from another tagging task. The proposed methods constitute pre-Training of a recurrent neural network or its parts to perform one tagging task given abundant and diverse training data, and then using it or its parts for a new task of tagging sound events of different nature, for which the data is limited. Several architectures for such transfer are proposed and evaluated, showing impressive classification accuracy of 83.4% with gains of up to 20 percentage points over the baseline given as little as 36 training samples for the target task.

U2 - 10.1109/WASPAA.2017.8169984

DO - 10.1109/WASPAA.2017.8169984

M3 - Conference contribution

SN - 978-1-5386-1631-4

T3 - IEEE Workshop on Applications of Signal Processing to Audio and Acoustics

SP - 6

EP - 10

BT - IEEE Workshop on Applications of Signal Processing to Audio and Acoustics

ER -