Convolutional Recurrent Neural Network and Data Augmentation for Audio
                        Tagging with Noisy Labels and Minimal Supervision

Ebbers, Janek; Häb‐Umbach, Reinhold

doi:10.33682/57xx-t679

Cited by 12 publications

(15 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is supported by the outcome of the contest-like comparison as well as the significance testing. This result reflects recent developments in the DCASE community, in which virtually all competition systems contain convolutional stages [21], [22], [25], [26], [28], [43]- [46]. However the paralinguistics community still relies primarily on pure recurrent networks [8], [9], [18], [19].…”

Section: Discussionsupporting

confidence: 64%

“…We chose this representation type because (1) it allows to use neural networks with two-dimensional inputs such as CNNs adapted from computer vision which currently prevail in audio recognition [32], and (2) it is currently the most prevalent spectrogram derivate for audio recognition, i.e. in the DCASE community [22], [24]- [28], [32] as well as paralinguistics [9], [16], [33].…”

Section: B Data Representationmentioning

confidence: 99%

“…applying the same activation function across the entire network, applying the same recurrent or convolutional type across the entire network, ensuring smooth unit growth across layers etc. [9], [16], [22], [24]- [28], [32], [33]. The drawback of this approach is that certain network configurations were excluded from the search space a priori which theoretically are valid according to our parent architecture scheme (see section II-C), such as CNNs with varying kernel shapes across the network.…”

Section: Hyperparameter Search Spacementioning

confidence: 99%

“…The main goal was to identify the best performing network type among the currently most prevailing ones, i.e. convolutional, recurrent, and fully-connected networks as well as combinations of thereof [22], [24]- [28]. More specifically, we aimed at (1) investigating how the arrangement and parametrization of well-known network components (i.e.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Comparison of Artificial Neural Network Types for Infant Vocalization Classification

Anders

Hlawitschka

Fuchs

2021

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

Section: Discussionsupporting

confidence: 64%

Section: B Data Representationmentioning

confidence: 99%

Section: Hyperparameter Search Spacementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Comparison of Artificial Neural Network Types for Infant Vocalization Classification

Anders

Hlawitschka

Fuchs

2021

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

“…During training various data augmentation techniques are used, namely random scaling, mixup, frequency warping, blurring, time masking, frequency masking and random noise. Random scaling and mixup [18] are performed on the waveform similar as in [19] by shifting and superposing signals as follows:…”

Section: Feature Extraction and Data Augmentationmentioning

confidence: 99%

Hidden Markov Model Variational Autoencoder for Acoustic Unit Discovery

et al. 2017

Self Cite

View full text Add to dashboard Cite

In this paper we present our system for the detection and classification of acoustic scenes and events (DCASE) 2020 Challenge Task 4: Sound event detection and separation in domestic environments. We introduce two new models: the forward-backward convolutional recurrent neural network (FBCRNN) and the tagconditioned convolutional neural network (CNN). The FBCRNN employs two recurrent neural network (RNN) classifiers sharing the same CNN for preprocessing. With one RNN processing a recording in forward direction and the other in backward direction, the two networks are trained to jointly predict audio tags, i.e., weak labels, at each time step within a recording, given that at each time step they have jointly processed the whole recording. The proposed training encourages the classifiers to tag events as soon as possible. Therefore, after training, the networks can be applied to shorter audio segments of, e.g., 200 ms, allowing sound event detection (SED). Further, we propose a tag-conditioned CNN to complement SED. It is trained to predict strong labels while using (predicted) tags, i.e., weak labels, as additional input. For training pseudo strong labels from a FBCRNN ensemble are used. The presented system scored the fourth and third place in the systems and teams rankings, respectively. Subsequent improvements allow our system to even outperform the challenge baseline and winner systems in average by, respectively, 18.0 % and 2.2 % event-based F1-score on the validation set.

show abstract

Emergence and function of cortical offset responses in sound termination detection

Sołyga¹,

Barkat²

2021

Preprint

View full text Add to dashboard Cite

Offset responses in auditory processing appear after a sound terminates. They arise in neuronal circuits within the peripheral auditory system, but their role in the central auditory system remains unknown. Here we ask what the behavioural relevance of cortical offset responses is and what circuit mechanisms drive them. At the perceptual level, our results reveal that experimentally minimizing auditory cortical offset responses decreases the mouse performance to detect sound termination, assigning a behavioural role to offset responses. By combining in vivo electrophysiology in the auditory cortex and thalamus of awake mice, we also demonstrate that cortical offset responses are not only inherited from the periphery but also amplified and generated de novo. Finally, we show that offset responses code more than silence, including relevant changes in sound trajectories. Together, our results reveal the importance of cortical offset responses in encoding sound termination and detecting changes within temporally discontinuous sounds crucial for speech and vocalization.

show abstract

Convolutional Recurrent Neural Network and Data Augmentation for Audio Tagging with Noisy Labels and Minimal Supervision

Cited by 12 publications

References 0 publications

Comparison of Artificial Neural Network Types for Infant Vocalization Classification

Comparison of Artificial Neural Network Types for Infant Vocalization Classification

Hidden Markov Model Variational Autoencoder for Acoustic Unit Discovery

Emergence and function of cortical offset responses in sound termination detection

Contact Info

Product

Resources

About