Handcrafted Feature From Classification Mood Music Indonesia With Machine Learning BERT and Transformer

Music is a combination of the human voice and instruments that bring beauty to the listener. Many people like music to create an atmosphere or atmosphere. In the field of computer science, with a lot of data, music is an object that can be studied through lyrics, audio, biographies and others. This research focuses on the mood of Indonesian music in the 70s and 80s by taking only the chorus. The features used in this study are chroma and spectrogram using handcrafted features consisting of pitch, harmonics, speech energy, pause, central moment. The data set used consists of text data and audio data. The mood tested consisted of sad, happy and neutral. This area of interest is investigated using the Crisp-DM method derived from business understanding, data understanding, data preparation, modeling, evaluation, deployment. Text processing using Bert. This study also uses the transformer model in machine learning. In constructing the Confusion Matrix, this study uses several parameters, namely epoch, dimensions, set size, learning rate and others. The value generated by the confusion matrix contains a good value of 53%.