Beyond the Dcase 2017 Challenge on Rare Sound Event Detection: A Proposal for a More Realistic Training and Test Framework

Baumann, J.; Lohrenz, Timo; Roy, Alexander; Fingscheidt, Tim

doi:10.1109/icassp40776.2020.9052950

Cited by 7 publications

(2 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The former was adapted from [14] modifications to fit the challenge task, while the latter deploys a one-dimensional CRNN, where only one frame context is used in the convolutional kernels to prevent temporal blur from the convolutional filtering. The SED-CRNN has a wide range of applications, where it was successfully employed in the past [3,[28][29][30], with the potential downside of not being task-specific enough for maximum performance. Instead, the 1D-CRNN is able to show very good taskspecific performance but only with very large amounts of training data and an optimized ensemble fusion approach.…”

Section: Related Workmentioning

confidence: 99%

A New DCASE 2017 Rare Sound Event Detection Benchmark Under Equal Training Data: CRNN With Multi-Width Kernels

Baumann

Meyer

Lohrenz

et al. 2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

Rare sound event detection (rare SED) deals with obtaining valuable information from data consisting mostly of acoustic background noises. It has meanwhile a long research history and was part of the DCASE 2017 Challenge. State-of-the-art performance is currently reached using a stacked combination of a CNN and an RNN, dubbed CRNN, which was also successfully applied in other domains such as in hybrid automatic speech recognition. In this work, we propose a new CRNN model for rare SED. This new model uses a set of parallel convolutions with multiple kernel widths in the CRNN and is based on an extended feature representation of the log-mel spectrogram. Furthermore, we apply and optimize different evaluation postprocessing methods and analyze the modifications in an ablation study. The proposed model outperforms the so-far top-scoring networks of the DCASE Challenge -using the same training material for all methods -by an error rate of 6.13 % absolute and by 4.39 % absolute in the F1 score on the test set and under these conditions achieves a new benchmark result on the DCASE 2017 Rare SED data set.

show abstract

Section: Related Workmentioning

confidence: 99%

A New DCASE 2017 Rare Sound Event Detection Benchmark Under Equal Training Data: CRNN With Multi-Width Kernels

Baumann

Meyer

Lohrenz

et al. 2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Since DCASE released Task 2 in 2017, rare sound event detection has received more and more attention [4][5][6]. In [7], Lim et al introduced a rare sound event detection system using the combination of one-dimensional (1D) convolutional neural network (1D ConvNet) and recurrent neural network (RNN) with long short-term memory units (LSTM), and ranked the first place in DCASE task 2.…”

Section: Introductionmentioning

confidence: 99%

Robust Sound Event Detection by a Two-Stage Network in the Presence of Background Noise

Liu

Zhou

et al. 2022

Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering

View full text Add to dashboard Cite

With the advent of deep learning, research on noise-robust sound event detection (SED) has progressed rapidly. However, SED performance in noisy conditions of single-channel systems remains unsatisfactory. Recently, there were several speech enhancement (SE) methods for the SED front-end to reduce the noise effect, which are completely two models that handle two tasks separately. In this work, we introduced a network trained by a two-stage method to simultaneously perform signal denoising and SED, where denoising and SED are conducted sequentially using neural network method. In addition, we designed a new objective function that takes into account the Euclidean distance between the output of the denoising block and the corresponding clean audio amplitude spectrum, which can better limit the distortion of the output features. The two-stage model is then jointly trained to optimize the proposed objective function. The results show that the proposed network presents a better performance compared with single-stage network without noise suppression. Compared with other recent state-of-the-art networks in the SED field, the performance of the proposed network model is competitive, especially in noisy environments.

show abstract

AI-Based, Automated Acoustic Diagnostics in Vehicles

Roy

2023

Proceedings

View full text Add to dashboard Cite

Beyond the Dcase 2017 Challenge on Rare Sound Event Detection: A Proposal for a More Realistic Training and Test Framework

Cited by 7 publications

References 12 publications

A New DCASE 2017 Rare Sound Event Detection Benchmark Under Equal Training Data: CRNN With Multi-Width Kernels

A New DCASE 2017 Rare Sound Event Detection Benchmark Under Equal Training Data: CRNN With Multi-Width Kernels

Robust Sound Event Detection by a Two-Stage Network in the Presence of Background Noise

AI-Based, Automated Acoustic Diagnostics in Vehicles

Contact Info

Product

Resources

About