Sirvi Autor "Raud, Karl" järgi

Nüüd näidatakse 1 - 1 1

Visual Piano Transcription
(Tartu Ülikool, 2025) Raud, Karl; Cabral Pinheiro, Victor Henrique, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Automatic music transcription (AMT) is a field that focuses on extracting symbolic representations from musical performances. Visual piano transcription (VPT) is a subproblem of AMT that uses only visual cues to transcribe piano performances. It is useful in cases where the audio is lost, noisy, or contains multiple instruments. In this work, an end-to-end convolutional deep learning approach for VPT is proposed, which predicts the keypresses of a piano performance, given a video of a person playing it. Three prior researches, including the current state of the art for VPT, were reimplemented under comparable conditions and evaluated against the proposed method on both an existing and a novel, out-of-distribution dataset compiled in the course of this study, to assess whether they can be used in real-world applications. The proposed method is shown to perform well under the tested conditions, surpassing the current state of the art. As a final set of evaluations, the current state of VPT is also directly compared to audio-based piano transcription (APT).