LibROSA Based Assessment Tool for Music Information Retrieval Systems

Music is possibly the most impactful bonding over the society and culture. The process of perusing the music in classrooms is considered as a costly affair for the humans in rural areas. Although the tutorials (i.e., video lectures) are usually available for free access in internet, the process of learning and evaluation yet depends on conventional teacher-student affair. So, the need for an automated tool designating the process of analysis cum assessment of music is formulated in the music eternity. This proposed model focuses on providing a solution for this use case exactly where the users can play notes or music pieces on a musical instrument (i.e., piano ) and followed by evaluating their performance against a chosen benchmark audio file, named as 'teacher' file. The technique considers various features such as loudness, tempo, rolloff frequency, kurtosis, skewness and centroid associated with piano for evaluation process. The model emphasizes towards to distinguish the characteristics of different keys and it was achieved with the help of Essentia onset function and LibROSA python package. The evaluation was carried through a set of stages such as normalization of features, pattern matching to extract an effective grading sheet of the played tune (i.e., 'Student' file). The drawbacks such as identification of the sequence of musical tones and noise onsets were avoided using a pattern matching framework, named as REMOVE_NOISY_ONSETS. The performance factor is fixed based on the notes that were missed by the user and the extra notes played by the user. Hence, this technique is of cutting-edge technology in the expertise of music education using technology trends. The proposed tool also helps people, those were deprived of learning the instrument of their choice by giving them an easy-to-use software tool to evaluate themselves and also give a lucid user interface to review their performance.