Unsupervised Transcription of Piano Music
Authors: Taylor Berg-Kirkpatrick, Jacob Andreas, Dan Klein
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our system outperforms the best published approaches on a standard piano transcription task, achieving a 10.6% relative gain in note onset F1 on real piano audio. and 4 Experiments |
| Researcher Affiliation | Academia | Taylor Berg-Kirkpatrick Jacob Andreas Dan Klein Computer Science Division University of California, Berkeley {tberg,jda,klein}@cs.berkeley.edu |
| Pseudocode | No | The paper describes models and update rules using mathematical equations and figures, but no explicit pseudocode blocks or algorithms are provided. |
| Open Source Code | No | The paper does not contain any statement about releasing source code or provide a link to a code repository. |
| Open Datasets | Yes | Data We evaluate on the MIDI-Aligned Piano Sounds (MAPS) corpus [14]. This corpus includes a collection of piano recordings from a variety of time periods and styles, performed by a human player on an acoustic Disklavier piano equipped with electromechanical sensors under the keys. ... and a large collection of symbolic music data from the IMSLP library [15, 16], used to estimate the event parameters in our model. ... [15] The international music score library project, June 2014. URL http://imslp.org. |
| Dataset Splits | Yes | In keeping with much of the existing music transcription literature, we use the first 30 seconds of each of the 30 ENSTDk Am recordings as a development set, and the first 30 seconds of each of the 30 ENSTDk Cl recordings as a test set. |
| Hardware Specification | No | The paper mentions the acoustic Disklavier piano used to generate the MAPS dataset recordings, but does not provide any specifications for the computational hardware (CPU, GPU, memory, etc.) used to run the experiments. |
| Software Dependencies | No | The paper describes methods like 'short-time Fourier transform' and mentions a 'commercially-available MIDI-to-sheet-music converter' but does not provide specific software names with version numbers for dependencies used in the experiments. |
| Experiment Setup | Yes | At decode time, to fit the spectral and envelope parameters and predict transcriptions, we run 5 iterations of the block-coordinate ascent procedure described in Section 3. |