Lele Liu

M.Sc. Lele Liu

Research Assistant / PostDoc

Computational Humanities

Universität Würzburg
Hubland Nord (ZPD)
Emil-Hilb-Weg 23

97074 Würzburg

Germany

Gebäude: ZPD

Raum: 02.007

E-Mail: lele.liu@uni-wuerzburg.de

Personal Homepage: https://cheriell.github.io

Research Interests

My main research interest is in music analysis. For the past years, I have been working on automatic music transcription, more broadly, in music information retrieval (e.g., beat tracking, music performance analysis, multimodal music analysis). Beyond music, I’m also curious about AI and language studies.

Education

09.2019–11.2025, PhD in Artificial Intelligence and Music, Queen Mary University of London, UK
01.2022–12.2022, Enrichment Scheme PhD Student, The Alan Turing Institute, UK
09.2017–09.2018, MSc in Sound and Music Computing, Queen Mary University of London, UK
09.2013–06.2017, BSc in Telecommunications Engineering with Management, Beijing University of Posts and Telecommunications, China & Queen Mary University of London, UK

Teaching

Seminar: Music Processing (MSc & MA), Winter 2025
Lecture: Music Information Retrieval (MSc & MA), Summer 2025
Seminar: Music Processing (MSc & MA), Winter 2024
Lecture: Music Information Retrieval (MSc & MA), Summer 2024
Seminar: Music Processing (MSc & MA), Winter 2023
Seminar: Music Transcription (MSc & MA), Winter 2023

Supervision

Master thesis: Michael May, "Synthetic-to-real domain transfer for automatic music transcription in orchestral music", 2024 to present
Master thesis: Moritz Hering, "Towards lightweight and causal neural networks for instrument-agnostic music transcription", 2024 to 2025
Master internship: Marc-Philipp Knechtle, "On the application of onsets and frames to classical music audio recordings", 2024
Bachelor internship: Markus Schuster, "Beat and downbeat tracking for classical music", 2025 to present
Bachelor thesis: Stefan Mühl, "An analysis of audiovisual features in movies across time and genre", 2023 to 2024

Publications

2025[ to top ]

Chiu, Ching‑Yu, Lele Liu, Christof Weiß, and Meinard Müller. “Cross‑Modal Approaches to Beat Tracking: A Case Study on Chopin Mazurkas”. Transactions of the International Society for Music Information Retrieval 8, no. 1 (2025): 55–69. https://doi.org/10.5334/tismir.238.
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ DOI ]
@article{Chiu_2025, author = {Chiu, Ching‑Yu and Liu, Lele and Weiß, Christof and Müller, Meinard}, journal = {Transactions of the International Society for Music Information Retrieval}, keywords = {myown}, number = 1, pages = {55–69}, publisher = {Ubiquity Press, Ltd.}, title = {Cross‑Modal Approaches to Beat Tracking: A Case Study on Chopin Mazurkas}, volume = 8, year = 2025 }

%0 Journal Article %1 Chiu_2025 %A Chiu, Ching‑Yu %A Liu, Lele %A Weiß, Christof %A Müller, Meinard %D 2025 %I Ubiquity Press, Ltd. %J Transactions of the International Society for Music Information Retrieval %N 1 %P 55–69 %R 10.5334/tismir.238 %T Cross‑Modal Approaches to Beat Tracking: A Case Study on Chopin Mazurkas %U http://dx.doi.org/10.5334/tismir.238 %V 8
Liu, Lele, and Christof Weiß. “Unsupervised Domain Adaptation for Music Transcription: Exploiting Cross-Version Consistency”. In ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 1-5. https://doi.org/10.1109/ICASSP49660.2025.10889474.
- [ BibTeX ]
- [ EndNote ]
- [ DOI ]
@inproceedings{10889474, abstract = {With the omnipresence of deep learning approaches in modern signal processing, the generalization of models to unseen domains is of crucial importance. Various strategies have been proposed to address such unsupervised domain adaptation including reconstruction autoencoders and adversarial training. In this paper, we propose another strategy using teacher–student learning, where we train a student model for the target domain using pseudo-labels generated by a teacher model trained on the source domain. To improve the adaptation, we propose to exploit cross-version data, i.e., target-domain data without labels but existing in different versions, which are expected to share the same labels. We process the pseudo-labels from the teacher model by keeping only consistent labels or by interpolating conflicting labels. In this paper, we demonstrate the proposed strategy for a music transcription subtask (multi-pitch estimation) using different performances of the same composition. We compare this strategy with a baseline method using a reconstruction autoencoder. Furthermore, we combine our strategy with the baseline method by enforcing the cross-version consistency of the target domain predictions. Our results show that our strategy helps to improve the unsupervised domain adaptation from one instrument (piano) to other instruments (singing and orchestra). Especially for the difficult transfer from piano music to complex orchestral music, we obtain substantial improvements using the proposed approach.}, author = {Liu, Lele and Weiß, Christof}, booktitle = {ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, keywords = {myown}, month = {04}, pages = {1-5}, title = {Unsupervised Domain Adaptation for Music Transcription: Exploiting Cross-Version Consistency}, year = 2025 }

%0 Conference Paper %1 10889474 %A Liu, Lele %A Weiß, Christof %B ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) %D 2025 %P 1-5 %R 10.1109/ICASSP49660.2025.10889474 %T Unsupervised Domain Adaptation for Music Transcription: Exploiting Cross-Version Consistency %X With the omnipresence of deep learning approaches in modern signal processing, the generalization of models to unseen domains is of crucial importance. Various strategies have been proposed to address such unsupervised domain adaptation including reconstruction autoencoders and adversarial training. In this paper, we propose another strategy using teacher–student learning, where we train a student model for the target domain using pseudo-labels generated by a teacher model trained on the source domain. To improve the adaptation, we propose to exploit cross-version data, i.e., target-domain data without labels but existing in different versions, which are expected to share the same labels. We process the pseudo-labels from the teacher model by keeping only consistent labels or by interpolating conflicting labels. In this paper, we demonstrate the proposed strategy for a music transcription subtask (multi-pitch estimation) using different performances of the same composition. We compare this strategy with a baseline method using a reconstruction autoencoder. Furthermore, we combine our strategy with the baseline method by enforcing the cross-version consistency of the target domain predictions. Our results show that our strategy helps to improve the unsupervised domain adaptation from one instrument (piano) to other instruments (singing and orchestra). Especially for the difficult transfer from piano music to complex orchestral music, we obtain substantial improvements using the proposed approach.

2024[ to top ]

Liu, Lele, and Christof Weiß. “Utilizing Cross-Version Consistency for Domain Adaptation: A Case Study on Music Audio”. In Tiny Papers Track at International Conference on Learning Representations.
- [ BibTeX ]
- [ EndNote ]
@inproceedings{liu2024utilizing, author = {Liu, Lele and Weiß, Christof}, booktitle = {Tiny Papers Track at International Conference on Learning Representations}, keywords = {myown}, title = {Utilizing cross-version consistency for domain adaptation: A case study on music audio}, year = 2024 }

%0 Conference Paper %1 liu2024utilizing %A Liu, Lele %A Weiß, Christof %B Tiny Papers Track at International Conference on Learning Representations %D 2024 %T Utilizing cross-version consistency for domain adaptation: A case study on music audio

2023[ to top ]

Bindi, Giovanni, Nils Demerlé, Rodrigo Diaz, David Genova, Aliénor Golvet, Ben Hayes, Jiawen Huang, et al. “AI (r)evolution -- Where Are We Heading? Thoughts about the Future of Music and Sound Technologies in the Era of Deep Learning”. https://arxiv.org/abs/2310.18320.
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
@misc{bindi2023airevolutionheading, author = {Bindi, Giovanni and Demerlé, Nils and Diaz, Rodrigo and Genova, David and Golvet, Aliénor and Hayes, Ben and Huang, Jiawen and Liu, Lele and Martos, Vincent and Nabi, Sarah and Pelinski, Teresa and Renault, Lenny and Sarkar, Saurjya and Sarmento, Pedro and Vahidi, Cyrus and Wolstanholme, Lewis and Zhang, Yixiao and Roebel, Axel and Bryan-Kinns, Nick and Giavitto, Jean-Louis and Barthet, Mathieu}, keywords = {myown}, title = {AI (r)evolution – where are we heading? Thoughts about the future of music and sound technologies in the era of deep learning}, year = 2023 }

%0 Generic %1 bindi2023airevolutionheading %A Bindi, Giovanni %A Demerlé, Nils %A Diaz, Rodrigo %A Genova, David %A Golvet, Aliénor %A Hayes, Ben %A Huang, Jiawen %A Liu, Lele %A Martos, Vincent %A Nabi, Sarah %A Pelinski, Teresa %A Renault, Lenny %A Sarkar, Saurjya %A Sarmento, Pedro %A Vahidi, Cyrus %A Wolstanholme, Lewis %A Zhang, Yixiao %A Roebel, Axel %A Bryan-Kinns, Nick %A Giavitto, Jean-Louis %A Barthet, Mathieu %D 2023 %T AI (r)evolution -- where are we heading? Thoughts about the future of music and sound technologies in the era of deep learning %U https://arxiv.org/abs/2310.18320

2022[ to top ]

Liu, Lele, Qiuqiang Kong, Veronica Morfi, and Emmanouil Benetos. “Performance MIDI-to-Score Conversion by Neural Beat Tracking”. In Proceedings of the International Society for Music Information Retrieval (ISMIR) Conference.
- [ BibTeX ]
- [ EndNote ]
@inproceedings{liu2022performance, author = {Liu, Lele and Kong, Qiuqiang and Morfi, Veronica and Benetos, Emmanouil}, booktitle = {proceedings of the International Society for Music Information Retrieval (ISMIR) Conference}, keywords = {myown}, title = {Performance MIDI-to-score conversion by neural beat tracking}, year = 2022 }

%0 Conference Paper %1 liu2022performance %A Liu, Lele %A Kong, Qiuqiang %A Morfi, Veronica %A Benetos, Emmanouil %B proceedings of the International Society for Music Information Retrieval (ISMIR) Conference %D 2022 %T Performance MIDI-to-score conversion by neural beat tracking

2021[ to top ]

Liu, Lele, and Emmanouil Benetos. “From Audio to Music Notation”. In Handbook of Artificial Intelligence for Music: Foundations, Advanced Approaches, and Developments for Creativity, edited by Eduardo Reck Miranda, 693-714. Cham: Springer International Publishing, 2021. https://doi.org/10.1007/978-3-030-72116-9_24.
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ DOI ]
@inbook{Liu2021, abstract = {The field of Music Information Retrieval (MIR) focuses on creating methods and practices for making sense of music data from various modalities, including audio, video, images, scores and metadata [54]. Within MIR, a core problem which to the day remains open is Automatic Music Transcription (AMT), the process of automatically converting an acoustic music signal into some form of musical notation. The creation of a method for automatically converting musical audio to notation has several uses including but also going beyond MIR: from software for automatic typesetting of audio into staff notation or other music representations, to the use of automatic transcriptions as a descriptor towards the development of systems for music recommendation, to applications for interactive music systems such as automatic music accompaniment, for music education through methods for automatic instrument tutoring, and towards enabling musicological research in sound archives, to name but a few.}, address = {Cham}, author = {Liu, Lele and Benetos, Emmanouil}, booktitle = {Handbook of Artificial Intelligence for Music: Foundations, Advanced Approaches, and Developments for Creativity}, editor = {Miranda, Eduardo Reck}, keywords = {myown}, pages = {693–714}, publisher = {Springer International Publishing}, title = {From Audio to Music Notation}, year = 2021 }

%0 Book Section %1 Liu2021 %A Liu, Lele %A Benetos, Emmanouil %B Handbook of Artificial Intelligence for Music: Foundations, Advanced Approaches, and Developments for Creativity %C Cham %D 2021 %E Miranda, Eduardo Reck %I Springer International Publishing %P 693--714 %R 10.1007/978-3-030-72116-9_24 %T From Audio to Music Notation %U https://doi.org/10.1007/978-3-030-72116-9_24 %X The field of Music Information Retrieval (MIR) focuses on creating methods and practices for making sense of music data from various modalities, including audio, video, images, scores and metadata [54]. Within MIR, a core problem which to the day remains open is Automatic Music Transcription (AMT), the process of automatically converting an acoustic music signal into some form of musical notation. The creation of a method for automatically converting musical audio to notation has several uses including but also going beyond MIR: from software for automatic typesetting of audio into staff notation or other music representations, to the use of automatic transcriptions as a descriptor towards the development of systems for music recommendation, to applications for interactive music systems such as automatic music accompaniment, for music education through methods for automatic instrument tutoring, and towards enabling musicological research in sound archives, to name but a few. %@ 978-3-030-72116-9
Liu, Lele, Veronica Morfi, and Emmanouil Benetos. “ACPAS: A Dataset of Aligned Classical Piano Audio and Scores for Audio-to-Score Transcription”. In International Society for Music Information Retrieval Conference Late-Breaking Demos Session (ISMIR-LBD).
- [ BibTeX ]
- [ EndNote ]
@conference{liu2021acpas, author = {Liu, Lele and Morfi, Veronica and Benetos, Emmanouil}, booktitle = {International Society for Music Information Retrieval Conference Late-Breaking Demos Session (ISMIR-LBD)}, keywords = {myown}, title = {ACPAS: A Dataset of aligned classical piano audio and scores for audio-to-score transcription}, year = 2021 }

%0 Generic %1 liu2021acpas %A Liu, Lele %A Morfi, Veronica %A Benetos, Emmanouil %B International Society for Music Information Retrieval Conference Late-Breaking Demos Session (ISMIR-LBD) %D 2021 %T ACPAS: A Dataset of aligned classical piano audio and scores for audio-to-score transcription
Liu, Lele, Veronica Morfi, and Emmanouil Benetos. “Joint Multi-Pitch Detection and Score Transcription for Polyphonic Piano Music”. In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 281-85. https://doi.org/10.1109/ICASSP39728.2021.9413601.
- [ BibTeX ]
- [ EndNote ]
- [ DOI ]
@inproceedings{9413601, abstract = {Research on automatic music transcription has largely focused on multi-pitch detection; there is limited discussion on how to obtain a machine- or human-readable score transcription. In this paper, we propose a method for joint multi-pitch detection and score transcription for polyphonic piano music. The outputs of our system include both a piano-roll representation (a descriptive transcription) and a symbolic musical notation (a prescriptive transcription). Unlike traditional methods that further convert MIDI transcriptions into musical scores, we use a multitask model combined with a Convolutional Recurrent Neural Network and Sequence-to-sequence models with attention mechanisms. We propose a Reshaped score representation that outperforms a LilyPond representation in terms of both prediction accuracy and time/memory resources, and compare different input audio spectrograms. We also create a new synthesized dataset for score transcription research. Experimental results show that the joint model outperforms a single-task model in score transcription.}, author = {Liu, Lele and Morfi, Veronica and Benetos, Emmanouil}, booktitle = {ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, keywords = {myown}, month = {06}, pages = {281-285}, title = {Joint Multi-Pitch Detection and Score Transcription for Polyphonic Piano Music}, year = 2021 }

%0 Conference Paper %1 9413601 %A Liu, Lele %A Morfi, Veronica %A Benetos, Emmanouil %B ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) %D 2021 %P 281-285 %R 10.1109/ICASSP39728.2021.9413601 %T Joint Multi-Pitch Detection and Score Transcription for Polyphonic Piano Music %X Research on automatic music transcription has largely focused on multi-pitch detection; there is limited discussion on how to obtain a machine- or human-readable score transcription. In this paper, we propose a method for joint multi-pitch detection and score transcription for polyphonic piano music. The outputs of our system include both a piano-roll representation (a descriptive transcription) and a symbolic musical notation (a prescriptive transcription). Unlike traditional methods that further convert MIDI transcriptions into musical scores, we use a multitask model combined with a Convolutional Recurrent Neural Network and Sequence-to-sequence models with attention mechanisms. We propose a Reshaped score representation that outperforms a LilyPond representation in terms of both prediction accuracy and time/memory resources, and compare different input audio spectrograms. We also create a new synthesized dataset for score transcription research. Experimental results show that the joint model outperforms a single-task model in score transcription.

2020[ to top ]

Liu, Lele, Veronica Morfi, and Emmanouil Benetos. “Joint Piano-Roll and Score Transcription for Polyphonic Piano Music”. In DMRN+15: Digital Music Research Network One-Day Workshop.
- [ BibTeX ]
- [ EndNote ]
@conference{liu2020joint, author = {Liu, Lele and Morfi, Veronica and Benetos, Emmanouil}, booktitle = {DMRN+15: Digital Music Research Network One-day Workshop}, keywords = {myown}, title = {Joint piano-roll and score transcription for polyphonic piano music}, year = 2020 }

%0 Generic %1 liu2020joint %A Liu, Lele %A Morfi, Veronica %A Benetos, Emmanouil %B DMRN+15: Digital Music Research Network One-day Workshop %D 2020 %T Joint piano-roll and score transcription for polyphonic piano music
Ycart, Adrien, Lele Liu, Emmanouil Benetos, and Marcus T. Pearce. “Musical Features for Automatic Music Transcription Evaluation”. https://arxiv.org/abs/2004.07171.
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
@misc{ycart2020musicalfeaturesautomaticmusic, author = {Ycart, Adrien and Liu, Lele and Benetos, Emmanouil and Pearce, Marcus T.}, keywords = {myown}, title = {Musical Features for Automatic Music Transcription Evaluation}, year = 2020 }

%0 Generic %1 ycart2020musicalfeaturesautomaticmusic %A Ycart, Adrien %A Liu, Lele %A Benetos, Emmanouil %A Pearce, Marcus T. %D 2020 %T Musical Features for Automatic Music Transcription Evaluation %U https://arxiv.org/abs/2004.07171
Ycart, Adrien, Lele Liu, Emmanouil Benetos, and Marcus T. Pearce. “Investigating the Perceptual Validity of Evaluation Metrics for Automatic Piano Music Transcription”. Transactions of the International Society for Music Information Retrieval 3, no. 1 (2020): 68–81. https://doi.org/10.5334/tismir.57.
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
- [ DOI ]
@article{Ycart_2020, author = {Ycart, Adrien and Liu, Lele and Benetos, Emmanouil and Pearce, Marcus T.}, journal = {Transactions of the International Society for Music Information Retrieval}, keywords = {myown}, number = 1, pages = {68–81}, publisher = {Ubiquity Press, Ltd.}, title = {Investigating the Perceptual Validity of Evaluation Metrics for Automatic Piano Music Transcription}, volume = 3, year = 2020 }

%0 Journal Article %1 Ycart_2020 %A Ycart, Adrien %A Liu, Lele %A Benetos, Emmanouil %A Pearce, Marcus T. %D 2020 %I Ubiquity Press, Ltd. %J Transactions of the International Society for Music Information Retrieval %N 1 %P 68–81 %R 10.5334/tismir.57 %T Investigating the Perceptual Validity of Evaluation Metrics for Automatic Piano Music Transcription %U http://dx.doi.org/10.5334/tismir.57 %V 3

2019[ to top ]

Liu, Lele, and Emmanouil Benetos. “Automatic Music Accompaniment With a Chroma-Based Music Data Representation”. In DMRN+14: Digital Music Research Network One-Day Workshop.
- [ BibTeX ]
- [ EndNote ]
@conference{liu2019automatic, author = {Liu, Lele and Benetos, Emmanouil}, booktitle = {DMRN+14: Digital Music Research Network One-day Workshop}, keywords = {myown}, title = {Automatic music accompaniment with a chroma-based music data representation}, year = 2019 }

%0 Generic %1 liu2019automatic %A Liu, Lele %A Benetos, Emmanouil %B DMRN+14: Digital Music Research Network One-day Workshop %D 2019 %T Automatic music accompaniment with a chroma-based music data representation

Hubland Nord, Geb. 23

M.Sc. Lele Liu

Bildnachweise