Search

We present a prompt-engineering-based text-augmentation approach applied to a language-queried audio source separation (LASS) task. To enhance the performance of LASS, the proposed approach utilizes large language models (LLMs) to generate multiple captions corresponding to each sentence of the training dataset. To this end, we first perform experiments to identify the most effective prompts for caption augmentation with a smaller number of captions. A LASS model trained with these augmented captions demonstrates improved performance on the DCASE 2024 Task 9 validation set compared to that trained without augmentation. This study highlights the effectiveness of LLM-based caption augmentation in advancing language-queried audio source separation., Comment: DCASE 2024 Challenge Task 9, 4 pages

This report describes the submission system by the GIST-AiTeR team for the VoxCeleb Speaker Recognition Challenge 2023 (VoxSRC-23) Track 4. Our submission system focuses on implementing diverse speaker diarization (SD) techniques, including ResNet293 and MFA-Conformer with different combinations of segment and hop length. Then, those models are combined into an ensemble model. The ResNet293 and MFA-Conformer models exhibited the diarization error rates (DERs) of 3.65% and 3.83% on VAL46, respectively. The submitted ensemble model provided a DER of 3.50% on VAL46, and consequently, it achieved a DER of 4.88% on the VoxSRC-23 test set., Comment: VoxSRC 2023 Track4

Deep neural networks (DNNs) have a high capacity to completely memorize noisy labels given sufficient training time, and its memorization, unfortunately, leads to performance degradation. Recently, virtual adversarial training (VAT) attracts attention as it could further improve the generalization of DNNs in semi-supervised learning. The driving force behind VAT is to prevent the models from overfitting data points by enforcing consistency between the inputs and the perturbed inputs. This strategy could be helpful in learning from noisy labels if it prevents neural models from learning noisy samples while encouraging the models to generalize clean samples. In this paper, we propose context-based virtual adversarial training (ConVAT) to prevent a text classifier from overfitting to noisy labels. Unlike the previous works, the proposed method performs the adversarial training at the context level rather than the inputs. It makes the classifier not only learn its label but also its contextual neighbors, which alleviates the learning from noisy labels by preserving contextual semantics on each data point. We conduct extensive experiments on four text classification datasets with two types of label noises. Comprehensive experimental results clearly show that the proposed method works quite well even with extremely noisy settings., Comment: 10 pages, Proceedings of Language Resources and Evaluation, 2022 (LREC'22)

Brain-computer interface uses brain signals to control external devices without actual control behavior. Recently, speech imagery has been studied for direct communication using language. Speech imagery uses brain signals generated when the user imagines speech. Unlike motor imagery, speech imagery still has unknown characteristics. Additionally, electroencephalography has intricate and non-stationary properties resulting in insufficient decoding performance. In addition, speech imagery is difficult to utilize spatial features. In this study, we designed length-wise training that allows the model to classify a series of a small number of words. In addition, we proposed hierarchical convolutional neural network structure and loss function to maximize the training strategy. The proposed method showed competitive performance in speech imagery classification. Hence, we demonstrated that the length of the word is a clue at improving classification performance., Comment: Submitted IEEE The 9th International Winter Conference on Brain-Computer Interface

Standoff chemical sensing of chemicals and toxic compounds is very desirable for actively monitoring residential, commercial, and industrial spaces. Analytes such as toxic industrial compounds (TICs), chemical warfare agents (CWAs), and environmental pollutants (EPs) can pose serious threats to civilians, warfighters, and the environment. We report on highly sensitive, robust, and inexpensive distributed standoff fiber optic chemical sensors (FOCSs). FOCSs are activated by analytes reaching a sensing region and inducing a change in its optical properties. The information of a spectral response is transmitted by a shift in transmission spectra probed by evanescent waves, corresponding to the detection of an analyte. However, there are many factors that play a role in the implementation and optimization of such fiber sensors. We find it beneficial to construct the theory of general optochemical waveguide sensors to guide the analysis. We apply it to a specific fiber platfom that is robust, broadband, multi-target, scalable, and compatible to texitle fibers and fabrics - a focus of this work. The good agreement between the predicted and measured spectra of FOCSs exposed to Ammonia (NH$_3$) lends support to the analysis. The experiment yields a 1ppm sensitivity, representing the current best in class. The use of NH$_3$ in the test serves as a representative for a larger set of amine-based TICs, CWAs, EPs, and other chemical species, targeted for wide-area distributed coverage and textile-compatible sensors.