Piano Transcription in the Studio Using an Extensible Alternating Directions Framework

Abstract

Given a musical audio recording, the goal of automatic music transcription is to determine a score-like representation of the piece underlying the recording. Despite significant interest within the research community, several studies have reported on a "glass ceiling" effect, an apparent limit on the transcription accuracy that current methods seem incapable of overcoming. In this paper, we explore how much this effect can be mitigated by focusing on a specific instrument class and making use of additional information on the recording conditions available in studio or home recording scenarios. In particular, exploiting the availability of single note recordings for the instrument in use we develop a novel signal model employing variable-length spectro-temporal patterns as its central building blocks -tailored for pitched percussive instruments such as the piano. Temporal dependencies between spectral templates are modeled, resembling characteristics of factorial scaled hidden Markov models (FS-HMM) and other methods combining Non-Negative Matrix Factorization with Markov processes. In contrast to FS-HMMs, our parameter estimation is developed in a global, relaxed form within the extensible alternating direction method of multipliers (ADMM) framework, which enables the systematic combination of basic regularizers propagating sparsity and local stationarity in note activity with more complex regularizers imposing temporal semantics. The proposed method achieves an f-measure of 93-95% for note onsets on pieces recorded on a Yamaha Disklavier (MAPS DB).

An augmented lagrangian method for piano transcription using equal loudness thresholding and lstm-based decoding

“…those close to the decision boundary. In [3], the corresponding threshold amin was derived from user input as this threshold depends on the recording level. More precisely, the user is asked to provide an example of a note having the lowest intensity to be expected in a recording session (during the evaluation this was one value for the entire dataset and not specific to recordings).…”

“…the activations for the onset part of each note pattern in P . The same representation was used in [3] for the final onset detection. We used LSTM networks in two different configurations.…”

“…However, for numerical reasons, the underlying parameter estimation process used in [10] was biased towards specific local minima of an objective function that are likely to cause misdetections. The design goal in [3] was thus to use a signal model similar to [10] but to replace the entire parameter estimation process. The resulting method consecutively switches from simple, convex regularizers (that stabilize the initial parameter estimation process) to more complex terms (to encourage a more meaningful structure as expressed by a graphical model).…”

“…However, such large, complex joint distributions have recently been successfully approximated using neural networks [18] [6]. Therefore, as a second extension, we investigate here combining the method proposed in [3], which is adaptable to new acoustical conditions with minimal effort, with long short term memory (LSTM) neural networks for decoding, which essentially provide a simple musical language model on top.…”

A Two-Stage Approach to Note-Level Transcription of a Specific Piano

“…In the supervised NMF, templates were usually formed by the isolated notes of the specific piano to be transcribed. Ewert employed spectro-temporal patterns to model the temporal evolution in NMF [31]. Cheng proposed a method to model the attack and decay of notes, and all the templates were trained by a Disklavier piano [32].…”