Jonghun Park

For effective style control in symbolic music generation, it is essential to understand and distinguish different musical styles. Although monophonic melody generation conditioned on musical style has gained attention, the question of whether or not monophonic melodies themselves carry sufficient stylistic information remains largely unexamined. In this paper, we explore the feasibility of classifying a musical genre solely based on symbolic monophonic melodies and propose a pattern-aware attention mask Transformer (PAMFormer) that integrates recurrent melodic patterns directly into the attention mechanism, guiding attention toward intra-pattern dependencies to capture genre-relevant motifs. Experiments conducted on two datasets demonstrate that PAMFormer consistently outperformed other architectures in genre classification of a monophonic melody. Furthermore, we explored the complementary strengths of pattern-based and sequence-based representations through the fusion of a pattern encoder and sequence encoder of PAMFormer. This hybrid approach yielded significant improvements in classification performance over individual representation baselines. Taken together, our findings highlight the effectiveness of incorporating recurrent patterns into the attention mechanism and also suggest the promise of combining pattern-based and sequence-based representations for symbolic monophonic melody understanding tasks. The source code and curated dataset used in this study are publicly available at: https://github.com/zZ-SMJ-Zz/PAMFormer-F.

Voice conversion is a task of synthesizing an utterance with target speaker's voice while maintaining linguistic information of the source utterance. While a speaker can produce varying utterances from a single script with different intonations, conventional voice conversion models were limited to producing only one result per source input. To overcome this limitation, we propose a novel approach for voice conversion with diverse intonations using conditional variational autoencoder (CVAE). Experiments have shown that the speaker's style feature can be mapped into a latent space with Gaussian distribution. We have also been able to convert voices with more diverse intonation by making the posterior of the latent space more complex with inverse autoregressive flow (IAF). As a result, the converted voice not only has a diversity of intonations, but also has better sound quality than the model without CVAE.

Automatic music transcription converts audio recordings into symbolic representations, facilitating music analysis, retrieval, and generation. A musical note is characterized by pitch, onset, and offset in an audio domain, whereas it is defined in terms of pitch and note value in a musical score domain. A time-aligned score, derived from timing information along with pitch and note value, allows matching a part of the score with the corresponding part of the music audio, enabling various applications. In this paper, we consider an extended version of the traditional note-level transcription task that recognizes onset, offset, and pitch, through including extraction of additional note value to generate a time-aligned score from an audio input. To address this new challenge, we propose an end-to-end framework that integrates recognition of the note value, pitch, and temporal information. This approach avoids error accumulation inherent in multi-stage methods and enhances accuracy through mutual reinforcement. Our framework employs tokenized representations specifically targeted for this task, through incorporating note value information. Furthermore, we introduce a pseudo-labeling technique to address a scarcity problem of annotated note value data. This technique produces approximate note value labels from existing datasets for the traditional note-level transcription. Experimental results demonstrate the superior performance of the proposed model in note-level transcription tasks when compared to existing state-of-the-art approaches. We also introduce new evaluation metrics that assess both temporal and note value aspects to demonstrate the robustness of the model. Moreover, qualitative assessments via visualized musical scores confirmed the effectiveness of our model in capturing the note values.

Automatic music transcription converts audio recordings into symbolic representations, facilitating music analysis, retrieval, and generation. A musical note is characterized by pitch, onset, and offset in an audio domain, whereas it is defined in terms of pitch and note value in a musical score domain. A time-aligned score, derived from timing information along with pitch and note value, allows matching a part of the score with the corresponding part of the music audio, enabling various applications. In this paper, we consider an extended version of the traditional note-level transcription task that recognizes onset, offset, and pitch, through including extraction of additional note value to generate a time-aligned score from an audio input. To address this new challenge, we propose an end-to-end framework that integrates recognition of the note value, pitch, and temporal information. This approach avoids error accumulation inherent in multi-stage methods and enhances accuracy through mutual reinforcement. Our framework employs tokenized representations specifically targeted for this task, through incorporating note value information. Furthermore, we introduce a pseudo-labeling technique to address a scarcity problem of annotated note value data. This technique produces approximate note value labels from existing datasets for the traditional note-level transcription. Experimental results demonstrate the superior performance of the proposed model in note-level transcription tasks when compared to existing state-of-the-art approaches. We also introduce new evaluation metrics that assess both temporal and note value aspects to demonstrate the robustness of the model. Moreover, qualitative assessments via visualized musical scores confirmed the effectiveness of our model in capturing the note values.

As the demand for multi-chip products with high capacity and small size increases, semiconductor packaging facilities have been faced with complicated constraints such as re-entrant flows, sequence dependent setups, and alternative routes, which leads to difficulties in scheduling semiconductor manufacturing operations. Furthermore, due to the frequent variations in the relative importance between objectives as well as the variabilities in initial setup status, available machines, and production requirements, practitioners are obliged to obtain a schedule within a short amount of computation time. In this paper, we propose a novel two-phase framework that aims to quickly produce a schedule of semiconductor packaging facilities by using case-based reasoning for minimizing the weighted sum of machine loss time and waiting time of jobs. Specifically, in the case generation phase, a case database is constructed by solving case scheduling problems using an existing solver. The case reasoning phase is responsible for repairing operation type sequences in the cases to produce a schedule for an unseen scheduling problem whose production requirements, available machines, initial setup status, and weight between performance measures are different from those of cases. The extensive experimental results demonstrated that the proposed approach requires a short computation time similar to the rule-based methods while maintaining the quality of the schedules comparable to that of the existing metaheuristics.