Abstract:
Teaching methods in music lessons are commonly structured involving instructors, who do not
offer students practicing individually timely feedback. This puts a negative twist on the method
since the students cannot follow or correct their learning in real-time. Current AMT systems
are incapable of offering feedback instantaneously, and they can only be applied to learning
environments that are diverse and learning levels as well. This project fills this gap by
implementing a real-time AMT system that transcribes notes correctly and provides the
feedback required for a great music learning experience.
To achieve real-time recognition of music notation for sounds, this system employs signal
processing combined with some criteria inherent to deep learning. First, the model of
transcription is learned from the volume of samples of a given set of instrumentations. Before
feeding the input in the model, there are a few pre-processing steps to be carried out, which
involve the elimination of noise and estimation of Pitch level.
A review of the initial performance of the AMT system shows a transcription with high
accuracy for different instruments and a processing with minimal latency, making the feedback
real-time. According to early testing carried out with music students and educators, the
integrated system results in a substantial improvement in practice efficiency by providing
practical feedback in real-time to inform the student of errors that need correction.
Subject Descriptors:
Applied computing-> Arts and humanities -> Sound and music computing
Computing methodologies -> Artificial intelligence-> Machine learning approaches -> Deep
learning
Keywords: Automatic Music Transcription, violin transcription, deep learning, CNN-LSTM,
music information retrieval, audio signal processing, feature extraction, music education, real-
time transcription, violin note recognition
