• Phonetics and Speech Sciences
• /
• v.15 no.3
• /
• pp.83-88
• /
• 2023

This study investigates the fine-tuning of large-scale Automatic Speech Recognition (ASR) models, specifically OpenAI's Whisper model, for domain-specific applications using the KsponSpeech dataset. The primary research questions address the effectiveness of targeted lexical item emphasis during fine-tuning, its impact on domain-specific performance, and whether the fine-tuned model can maintain generalization capabilities across different languages and environments. Experiments were conducted using two fine-tuning datasets: Set A, a small subset emphasizing specific lexical items, and Set B, consisting of the entire KsponSpeech dataset. Results showed that fine-tuning with targeted lexical items increased recognition accuracy and improved domain-specific performance, with generalization capabilities maintained when fine-tuned with a smaller dataset. For noisier environments, a trade-off between specificity and generalization capabilities was observed. This study highlights the potential of fine-tuning using minimal domain-specific data to achieve satisfactory results, emphasizing the importance of balancing specialization and generalization for ASR models. Future research could explore different fine-tuning strategies and novel technologies such as prompting to further enhance large-scale ASR models' domain-specific performance.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.24 no.4
• /
• pp.382-391
• /
• 2021

Unsupervised domain adaptation often gives impressive solutions to handle domain shift of data. Most of current approaches assume that unlabeled target data to train is abundant. This assumption is not always true in practices. To tackle this issue, we propose a general solution to solve the domain gap minimization problem without any target data. Our method consists of two regularization steps. The first step is a pixel regularization by arbitrary style transfer. Recently, some methods bring style transfer algorithms to domain adaptation and domain generalization process. They use style transfer algorithms to remove texture bias in source domain data. We also use style transfer algorithms for removing texture bias, but our method depends on neither domain adaptation nor domain generalization paradigm. The second regularization step is a feature regularization by feature alignment. Adding a feature alignment loss term to the model loss, the model learns domain invariant representation more efficiently. We evaluate our regularization methods from several experiments both on small dataset and large dataset. From the experiments, we show that our model can learn domain invariant representation as much as unsupervised domain adaptation methods.

• Bulletin of the Korean Mathematical Society
• /
• v.49 no.1
• /
• pp.205-211
• /
• 2012

Let R be a graded Noetherian domain and A a graded Krull overring of R. We show that if h-dim $R\leq2$, then A is a graded Noetherian domain with h-dim $A\leq2$. This is a generalization of the well-know theorem that a Krull overring of a Noetherian domain with dimension $\leq2$ is also a Noetherian domain with dimension $\leq2$.

• Journal of the Korean Mathematical Society
• /
• v.51 no.6
• /
• pp.1209-1220
• /
• 2014

We prove global gradient estimates in weighted Orlicz spaces for weak solutions of nonlinear elliptic equations in divergence form over a bounded non-smooth domain as a generalization of Calder$\acute{o}$n-Zygmund theory. For each point and each small scale, the main assumptions are that nonlinearity is assumed to have a uniformly small mean oscillation and that the boundary of the domain is sufficiently flat.

• The Journal of Society for e-Business Studies
• /
• v.7 no.3
• /
• pp.49-60
• /
• 2002

Many researchers have studied component technologies with concept, methodology and implementation for partial business domain, however there are rarely researches for component classification to manage these systematically. In this paper, we suggest a component classification model, which can make component reusability higher and can derive higher productivity of software development. We take four focuses generalization, abstraction, technology and size. The generalization means which category a component belongs to. The abstraction means how specific a component encapsulates its inside. The technology means which platform for hardware environment a component can be plugged in. The size means the physical component volume.

• Journal of the Korean Mathematical Society
• /
• v.58 no.1
• /
• pp.149-171
• /
• 2021

Let Cl(A) denote the class group of an arbitrary integral domain A introduced by Bouvier in 1982. Then Cl(A) is the ideal class (resp., divisor class) group of A if A is a Dedekind or a Prüfer (resp., Krull) domain. Let G be an abelian group. In this paper, we show that there is a ring of Krull type D such that Cl(D) = G but D is not a Krull domain. We then use this ring to construct a Prüfer ring of Krull type E such that Cl(E) = G but E is not a Dedekind domain. This is a generalization of Claborn's result that every abelian group is the ideal class group of a Dedekind domain.

• Steel and Composite Structures
• /
• v.46 no.3
• /
• pp.435-449
• /
• 2023

Time domain method and frequency domain method are commonly used in the current fatigue life calculation theory. The time domain method has complicated procedures and needs a large amount of calculation, while the frequency domain method has poor applicability to different materials and different spectrum, and improper selection of spectrum model will lead to large errors. Considering that artificial neural network has strong ability of nonlinear mapping and generalization, this paper applied this technique to random fatigue life prediction, and the effect of average stress was taken into account, thereby achieving more accurate prediction result of random fatigue life.

• Honam Mathematical Journal
• /
• v.38 no.3
• /
• pp.613-623
• /
• 2016

As a generalization of the Fourier transform, the fractional Fourier transform plays an important role both in theory and in applications of signal processing. We present a new approach to reach a uniform sampling theorem in the shift-invariant spaces associated with the fractional Fourier transform domain.

• Bulletin of the Korean Mathematical Society
• /
• v.51 no.6
• /
• pp.1669-1687
• /
• 2014

We consider a nonuniformly nonlinear elliptic systems with resonance part and nonlinear Neumann boundary condition on an unbounded domain. Our arguments are based on the minimum principle and rely on a generalization of the Landesman-Lazer type condition.

• Journal of the Korean Mathematical Society
• /
• v.57 no.4
• /
• pp.973-986
• /
• 2020

In this paper, we compute the Hankel matrix representation of the Hankel operator on the Hardy space of a general bounded domain with respect to special orthonormal bases for the Hardy space and its orthogonal complement. Moreover we obtain the compact form of the Hankel matrix for the unit disc case with respect to these bases. One can see that the Hankel matrix generated by this computation turns out to be a generalization of the case of the unit disc from the single simply connected domain to multiply connected domains with much diversities of bases.